remote sensing applications for carbonatite …€¦ · carbonatites, alkali volcanics,...

iv

REMOTE SENSING APPLICATIONS FOR CARBONATITE

ASSESSMENT AND MAPPING USING VNIR AND SWIR BANDS AT AL-

UYAYNAH, UA E

By

Ali M. Assiri

A DISSERTATION

Submitted to

Michigan State University

in partial fulfillment of the requirements

for the degree of

Geological Sciences-Doctor of philosophy

2016

i

ABSTRACT

REMOTE SENSING APPLICATIONS FOR CARBONATITE ASSESSMENT AND

MAPPING USING VNIR AND SWIR BANDS

By

Ali M. Assiri

This dissertation undertakes the experimental analysis of multispectral reflectance data to

remotely differentiate and lithologically map the extrusive carbonatites from the adjoining rock

units in the Al-Uyaynah region located within the northeastern terrain of UAE. The study employs

visible/near-infrared (VNIR), shortwave-infrared (SWIR) spectral measurements developed in the

laboratory using ASD, Landsat OLI, and the Advanced Spaceborne Thermal Emission and

Reflection Radiometer (ASTER) images of extrusive carbonatites in Al-Uyaynah metamorphic

sole units. Chapter 1 of the study is a succinct overview of the entire effort. It also encompasses

the context of remote sensing and the use of the reflected spectrum from visible to shortwave-

infrared in mineral exploration. In particular I focus on carbonatite identification and analysis for

the purpose of geological mapping. In Chapter 2 of this study, I perform tests and analyses on new

Landsat 8 image data over Al-Uyaynah- UAE so as to examine the validitity of this system for

medium resolution distantly sensed reflectance measurements. Specifically, I examine the utility

of Landsat 8 imagery to define carbonatite outcrops and relevant lithological components. Chapter

3 of the study examines the Thermal Emission and Reflection Radiometer (ASTER) VNIR and

SWIR data over the Al-Uyaynah region. In this Chapter I leveraged an existing low resolution

geologic map, ASD laboratory measurements, and field data to examine the spatial distrubtion of

carbonatites, alkali volcanics, metavolcanics, and an ophiolite complex using ASTER imagery. I

further tested the efficency of four different mapping algorithms for their utility in effectively

ii

discriminating between the units and their locations. Chapter 4 of my dissertation examines the

relationship between the composition and spectral charachteristics of carbonatite rocks. In this

Chapter, carbonatite samples were examined spectrally within the visible near infrared (VNIR)

and short wave infrared (SWIR) bands using a lab-based spectroradiometer device (ASD). The

carbonatite rocks samples, were analyzed geochemically to determine the major oxides

components in the weight percentage (wt%) using X-ray fluorescence (XRF). Inductively Coupled

Plasma-Mass Spectrometer (ICP-MS) techniques were then applied in order to determine the rare

earth elements (REEs) concentrations in the carbonatites. The ASD spectra and the chemical

analysis results offered us a deep understading of the connection between the rock composition

and spectral behavior of Al-Uyaynah carbonatite rocks. In Chapter 5 of this study, the entire

research effort is combined in a synthesis that links the outcomes from each chapter into a

comprehensive study of the spectral characteristics of carbonatites from the Al-Uyaynah region.

Pathways for further of research are also recommended. This investigation is significant as it

presents: (1) New ways of employing multispectral systems, Landsat 8 and ASTER, for use in the

geological mapping of carbonatites; (2) Methods to recognize the mineralogical indicators for

ferro/calcio-carbonatites; (3) evalutions of mapping algorithms and creates rankings based on their

precision and accuracy for utilization in complex metamorphosed geologic terraines.

iv

ACKNOWLEDGEMENTS

This effort has materialized through quite a few collaborations that I am so honored to have

developed through my Ph.D. I feel deeply indebted and acknowledge the invaluable contributions

of others to this study. First of all, my gratitude is due to my research advisor, Dr. Tyrone O.

Rooney, for his commitment and for accepting me in his research group, as well as for his valuable

support and mentorship extended to me throughout my Ph.D. Thank you, Dr. Rooney, for

transforming me so positively. In addition, many thanks are due to Dr. Richard Becker, my

research committee member, for helping me to expand my professional experience of remote

sensing and mineral spectroscopy, for his invaluable support in laboratory spectral measurement

and analysis. Dr. Jiaguo Qi is another person deserving deep appreciation because of the time he

devoted in teaching me advanced applications of remote sensing. Dr. Thomas Vogel also deserves

my gratefulness for his time and encouragement. I would like thank Neal Langford for his proof-

reading and outstanding editing of chapter 4 of this dissertation.

My special gratitude is due to the Department of Geological Science at Michigan State

University staff for being so professionally devoted to keeping the system going so smoothly for

the growth of their students. Dr. Rooney and his research fellows (Chris, Guillaume) deserve my

humble thanks for assisting me with the sample preparations and for providing me with the

practical support for the XRF procedures, for devoting their time on the South Campus dealing

with ICP-MS instruments. Your support has been instrumental!

v

Many thanks are extended to Dr. Baha-Eldeen for helping me discover the geology of Al-

Uyaynah-UAE, for managing his schedule to visit the field with me. Also would like to thank the

Geology Department staff at United Arab Emirates University for providing the logistic supports.

Similarly, I would also want to acknowledge the support of NASA and Land Processes

Distributed Active Archive Center (LP DAAC), USGS Earth Resources Observation and Science

(EROS) Center in Sioux Falls, South Dakota for awarding me the imagery grant to invaluably help

me during my study

KACST deserves my respect and appreciation for funding my education tuitions, stipend,

funding to attend conferences. Without this grant, my research would not have become a material

reality.

vi

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................................... ix

LIST OF FIGURES ...................................................................................................................... x

CHAPTER 1 ................................................................................................................................. 1

DISSERTATION PHASES AND BACKGROUND .................................................................. 1

1.Introduction ...................................................................................................................... 1

2.Carbonatite ....................................................................................................................... 5

2.1.Carbonatite definition and subtypes ......................................................................... 5

2.2 Carbonatite petrogenesis .......................................................................................... 8

2.3 Rocks spatially connected with carbonatites............................................................ 9

2.4 Carbonatite tectonic settings .................................................................................. 11

2.5 Economic significance of carbonatite .................................................................... 13

3.Detecting, assessing, and analyzing carbonatites and surrounding rock units at Al-

Uyaynah, UAE using remote sensing techniques. ............................................................ 17

4.Understanding carbonatite chemistry using spectral analysis devices (ASD) and

geochemical approaches. .................................................................................................. 22

5.Dissertation phases......................................................................................................... 22

REFERENCES ................................................................................................................. 24

CHAPTER 2 ................................................................................................................................ 30

CARBONATITE IDENTIFICATION USING LANDSAT 8 VNIR-SWIR BANDS AT AL-

UYAYNAH, UAE ........................................................................................................................ 30

Abstract ............................................................................................................................. 30

1.Introduction .................................................................................................................... 31

2.Al-Uyaynah .................................................................................................................... 33

3.Methods.......................................................................................................................... 34

3.1.Landsat 8 data calibration ...................................................................................... 34

3.2.Landsat data analysis.............................................................................................. 37

3.3.Field work and ASD spectral measurements ......................................................... 37

3.4.The spectral angle mapper (SAM) ......................................................................... 38

4.Results ............................................................................................................................ 38

4.1.The image processing results ................................................................................. 38

4.2.Carbonatite endmember spectral characteristics .................................................... 42

4.2.1.The ferro-carbonatite  ..................................................................................... 42

4.2.2.The calcio-carbonatite  ................................................................................... 44

4.2.3.The difference between ferro/calcio-carbonatite endmembers ...................... 46

4.2.4.Other endmembers ......................................................................................... 46

4.3.The SAM classification results .............................................................................. 49

5.Discussion ...................................................................................................................... 49

5.1.Distribution of calcio/ferro-carbonatites units ....................................................... 53

5.2.Distribution of other non-carbonatite units ............................................................ 54

6.Conclusion ..................................................................................................................... 55

vii

APPENDIX ....................................................................................................................... 57

REFERENCES ................................................................................................................. 59

CHAPTER 3 ............................................................................................................................... 63

A COMPARISON OF MAPPING METHODS AND SPECTRAL ANALYSIS TO

IMPROVE CARBONATITE MAPPING USING VNIR AND SWIR BANDS OF

ASTER DATA AT AL-UYAYNAH AREA, UAE ................................................................... 63

Abstract ............................................................................................................................. 63

1.Introduction .................................................................................................................... 64

2.Geologic background of study ....................................................................................... 68

2.1 Carbonatite location and outcrops .......................................................................... 68

2.2 Geochemistry of Al-Uyaynah Carbonatite ............................................................ 71

3. Methods......................................................................................................................... 71

3.1 ASTER data calibration ......................................................................................... 71

3.2 ASTER data analysis.............................................................................................. 72

3.2.1 Dimensionality reduction ............................................................................... 72

3.2.2 Endmember extraction ................................................................................... 75

3.2.3 Defining the regions of interest (ROIs) .......................................................... 75

3.3 Classification methods ........................................................................................... 77

3.3.1 The Binary Encoding Classification (BEC) ................................................... 79

3.3.2 The Spectral Information Divergence (SID) .................................................. 79

3.3.3 Spectral Angle Mapper (SAM) ...................................................................... 80

3.3.4 Support Vector Machines (SVM) .................................................................. 82

3.4 Post classification assessments .............................................................................. 84

4. Field surveys and ASD measurements ......................................................................... 86

5. Results ........................................................................................................................... 87

5.1 Noise reduction and endmembers extraction ......................................................... 87

5.2 Endmembers' spectral characteristics ..................................................................... 91

5.2.1 Calcite ............................................................................................................. 91

5.2.2 Alkali amphibole minerals ............................................................................. 92

5.2.3 Iron minerals .................................................................................................. 92

5.2.4 REE minerals .................................................................................................. 93

5.3 Classification and accuracy assessment ................................................................. 94

5.3.1 Binary Encoding Classification (BEC) .......................................................... 96

5.3.2 Spectral Information Divergence (SID) ......................................................... 96

5.3.3 Spectral Angle Mapper (SAM) ...................................................................... 96

5.3.4 Support Vector Machine (SVM) .................................................................... 97

6. Discussion ..................................................................................................................... 97

7. Testing SVM as a carbonatite detection tool - performance at an alternate locality .. 108

8. Conclusion .................................................................................................................. 110

APPENDIX ..................................................................................................................... 113

REFERENCES ............................................................................................................... 121

CHAPTER 4 .............................................................................................................................. 127

viii

DERIVING THE GEOCHEMICAL CHARACTERISTIC OF CARBONATITES

USING VISIBLE NEAR INFRARED AND SHORT WAVE INFRARED

SPECTROSCOPY .................................................................................................................... 127

Abstract ........................................................................................................................... 127

1.Introduction .................................................................................................................. 129

2.Background .................................................................................................................. 130

3.Study area location and geology .................................................................................. 134

4.Methods........................................................................................................................ 137

4.1.Sample preparation............................................................................................... 137

4.2.Spectral Analysis using ASD ............................................................................... 137

4.3.Geochemical analysis ........................................................................................... 138

5.Results .......................................................................................................................... 141

5.1.Carbonatite Spectra .............................................................................................. 141

5.1.1.Calcite related absorption ............................................................................. 141

5.1.2.Iron related absorption ................................................................................. 144

5.1.3.Rare earth element (REE) related absorption ............................................... 145

5.1.4.Major element analysis by XRF ................................................................... 150

5.2.Geochemical impact of weathering on major element composition .................... 151

5.3.CIPW normative mineral composition................................................................. 153

5.4.Trace element analysis by LA-ICPMS ................................................................ 154

6.Discussion .................................................................................................................... 162

6.1.The relationship between bulk composition and carbonatite spectra ................... 162

6.2.The spectra of a carbonatite weathered surface ................................................... 169

6.3.Nd spectra absorption features in carbonatite rocks ............................................ 170

7.Conclusion ................................................................................................................... 172

REFERENCES ............................................................................................................... 175

CHAPTER 5 .............................................................................................................................. 180

CONCLUSION AND RECOMMENDATIONS .................................................................... 180

1.Dissertation Synthesis .................................................................................................. 180

2.Dissertation conclusion ................................................................................................ 181

3.Recommendations and outlooks for carbonatite prospectors….. ................................ 183

ix

LIST OF TABLES

Table 1.1. Occurrence of rocks accompanying carbonatites [Count is number of deposits; % is

percentage of deposits with related rock type, data were acquired from USGS Open-File Report

[1]. ................................................................................................................................................. 19

Table 1.2. Shows the rare earth element (REE)- Major bearing minerals [9] .............................. 20

Table 2.1. The values of threshold that used for dividing the results of statistical kappa

coefficient degrees of agreement. From [3-5]. The values close or equal to 1.0 specify perfect

agreement; while any values between 0.85-0.99 show excellent agreement, and any kappa

coefficient values of between 0.75-0.85 would be a very good agreement. The lesser degree of

agreement were classified as: good if the kappa coefficient values were between 0.4-0.55; fair

agreement between 0.40-0.55 values; and kappa coefficient values between 0.4-0.05 or fewer

show poor to no agreement. The SAM map accuracy assessment displays overall accuracy with

66.92%. Kappa Coefficient is 0.58, which falls into the ‘good’ category. ................................... 50

Table 3.1. The values of threshold that used for dividing the results of statistical kappa

coefficient degrees of agreement. From [52, 57]. ....................................................................... 103

Table 4.1. Illustrates the major oxides distribution of selected Al-Uyaynah carbonatite samples

evaluated using XRF. All carbonatite samples contain calcium concentrations 26-45 Wt. %. The

lower concentrations of carbonate were recorded in ferro-carbonatites fresh and weathered

samples and those ferro-carbonatites exhibit a higher abundance of iron and silica in comparison

to calcio-carbonatites. ................................................................................................................. 155

Table 4.2. Shows the distributions of REEs of selected carbonatite field samples from Al-

Uyaynah field. The REE (ppm) were analyzed by (LA-ICP-MS).............................................. 156

Table 4.3. Shows the CIPW normative mineral compositions were calculated for Al-Uyaynah

carbonatite samples from the whole rock XRF geochemical data. ............................................. 157

x

LIST OF FIGURES

Figure 1.1. Illustration of the ranges of the electromagnetic spectrum showing the most crucial

ranges for mineral sensing spectroscopy applications - the visible/near infrared (VNIR), and

short-wave infrared (SWIR) [10]. ................................................................................................... 4

Figure 1.2. Global distribution of Nb– and REE–bearing carbonatite deposits [6]. ....................... 7

Figure 1.3. Classification diagram of carbonatite based on its chemistry; modified from [12]. .... 7

Figure 1.4. Displays the tectonic setting of Al-Uyaynah area within Arabian plat. (modified

from [7]). ....................................................................................................................................... 10

Figure 1.5. Map of the northern eastern of UAE indicating the location of the Al-Uyaynah area

and general geological units left, modified from [1, 2]. (b) ASTER 3,2,1 natural color image

right. .............................................................................................................................................. 12

Figure 1.6. Rare elements (REEs) and selected high field strength element distribution in the

mantle and earth’s crust, which is normalized to CI chondrite [9]. .............................................. 15

Figure 1.7. Geological map of Al-Uyaynah area showing the extrusive carbonatites (modified

from [8]). ....................................................................................................................................... 16

Figure 1.8. A comparison between the occurrence of rocks accompanying carbonatites [Count is

number of deposits; % is percentage of deposits with related rock type, data used in this diagram

were obtained from table3 of USGS Open-File Report [6] .......................................................... 18

Figure 2.1. A map of the northern eastern of UAE indicating the location of the Al-Uyaynah area

and general geological units, modified from [1, 2], and Landsat 8 natural colored image of Al-

Uyaynah area shows the location of study area (right). ................................................................ 35

Figure 2.2. A field view of Al-Uyaynah carbonatites (fig. 1), (a) shows the ferro-carbonatites

outcrops of the northern end of the study area, (b, and c) show a close view of ferro-carbonatites.

(d) Shows the extruded calcio-carbonatite, and (e, f) illustrates a close view of calcio-carbonatite.

....................................................................................................................................................... 36

Figure 2.3. MNF eigenvalue plot for study area (Al-Uyaynah). All MNF bands have eigenvalues

values greater than 1. The lowest band (7) has eigenvalue equal to 1.23, which means it contains

coherent data. However, the first three bands contain more spatially coherent information than

the others. ...................................................................................................................................... 39

xi

Figure 2.4. MNF images of Landsat 8 VNIR-SWIR bands for study area. Images 1-7, display a

single gray band, while image 8 shows a combination of first three MNF bands (1, 2 and 3 as

RGB). Image 9 shows natural color image (Landsat Bands 3, 2, 1 (RGB)) for comparison (Not

MNF image). ................................................................................................................................. 40

Figure 2.5 Show a 3-D Scatter plot of the end-members data cloud (image bands 1-3 were used

as plot axes). The identical rock units were characterized by the same color. ............................. 41

Figure 2.6. ASD Resampled spectra to Landsat 8 of (a) ferro-carbonatite and (b) calcio-

carbonatite. .................................................................................................................................... 43

Figure 2.7. A laboratory reflectance spectra (ASD) of samples, (a) ferro-carbonatite and (b)

calcio-carbonatite (left). (a) A zoom of spectral region 0.5-0.6 µm of laboratory reflectance

spectra (ASD) of REE absorption are near 0.589, 0.567, and 0.56 µm, and (b) A zoom of spectral

region 2000-2400 nm of laboratory reflectance spectra (ASD) calcite (CO3, MgO) REE

absorption are near 2.32-2.35µm and calcite has weak absorption at 1.6 µm. ............................. 43

Figure 2.8. Landsat 8 image endmembers spectra of (a) calcio-carbonatite in red color and (b)

ferro-carbonatite in green color; within VNIR-SWIR bands. ....................................................... 45

Figure 2.9. The image n-D spectra were extracted from Landsat OLI image used in SAM

classification. ................................................................................................................................ 48

Figure 2.10. SAM n-D classification map of Al-Uyaynah area shows the extrusive carbonatites

and the lithology unites. ................................................................................................................ 51

Figure 2.11. Geological map of Al-Uyaynah area shows the extrusive carbonatites (modified

from British geological survey, 2006). ......................................................................................... 52

Figure 2.12. A Comparison between the rock units’ distribution that classified correctly, in

percent, of SAM map based on the results of the confusion matrix accuracy assessment. SAM

method gave, overall, a good result after comparing SAM map to the field ground truth regions.

....................................................................................................................................................... 58

Figure 3.1. ASTER spectral transmittance of band pass filters were used in this research,

modified from[13]. ........................................................................................................................ 65

Figure 3.2. (a) Map of the northern eastern of UAE indicating the location of the Al-Uyaynah

area and general geological units, modified from [23, 24]. (b) ASTER 3, 2, and 1 natural color

image. ............................................................................................................................................ 69

Figure 3.3. A field view of Al-Uyaynah carbonatites (fig. 1), (a) shows the ferro-carbonatites


xii

(d) Shows the extruded calcio-carbonatite, and (e, f) illustrates a close view of calcio-carbonatite.

....................................................................................................................................................... 70

Figure 3.4. A MNF eigenvalue plot for study area (Al-Uyaynah). All MNF bands have

eigenvalues values greater than 1. The lowest band (9) has eigenvalue equal to 2.53, which

means it contains coherent data. However, the first 5 bands contain more spatially coherent

information than the other bands. ................................................................................................. 73

Figure 3.5 MNF images of ASTER VNIR-SWIR bands for study area. Images 1-9, display a

single gray band, while image 10 shows a combination of first three MNF bands (1, 2 and 3 as

RGB) for comparison .................................................................................................................... 74

Figure 3.6. Show a 3-D Scatter plot of the end-members data cloud (image bands 1-3 were used

as plot axes). The identical rock units were characterized by the same color. ............................. 76

Figure 3.7. Shows a random sampling of pixels that were generated from the defined ROIs.

These random pixels were used as a reference to assess the mapping outcomes of the BEC, SID,

SAM, and SVM methods. ............................................................................................................. 78

Figure 3.8. Shows how the SAM method classifies the image. The smaller the angles between

the endmember and pixel, the more of a match there exists with the endmember spectrum. Where

pixels surpass the defined angle threshold, they are not categorized. The figure was modified

from [37]. ...................................................................................................................................... 81

Figure 3.9. Linear Support Vector Machine illustration. The SVM training algorithm is used to

find a hyperplane that divides the image pixels to a separate number of classes. Modified from

[41, 42]. ......................................................................................................................................... 83

Figure 3.10. Laboratory reflectance spectra (ASD), showing REE absorption near 0.589, 0.567,

and 0.56 µm and (CO3, MgO) absorption near 2.32-2.35µm, and near 1.6 µm for calcite

absorption. ..................................................................................................................................... 88

Figure 3.11. The resampled laboratory reflectance spectra (ASD) of ferro-carbonatite and calcio-

carbonatite. .................................................................................................................................... 88

Figure 3.12. A zoom-in of the spectral region 0.5-0.6 µm of laboratory reflectance spectra

(ASD), showing REE absorption near 0.589, 0.567, and 0.56 µm (left), and a zoom-in of the

spectral region 0.2-2.4 µm of laboratory reflectance spectra (ASD) showing (CO3, MgO)

absorption near 2.32-2.35µm, and near 1.6 µm for calcite absorption. ........................................ 89

Figure 3.13. ASTER image endmemebers spectra of calcio-carbonatite and ferro-carbonatite; in

VNIR-SWIR bands. ...................................................................................................................... 90

xiii

Figure 3.14. The image endmember spectra were extracted from Landsat OLI image and are

utilized as endmembers for further image classification. ............................................................. 95

Figure 3.15. The binary encoding classification (BEC) map shows very poor discrimination

among ferro/calcio-carbonatite and other the rock units. The accuracy assessment of the overall

accuracy is 39%, the Kappa Coefficient is 0.31, this recognized as "No agreement" in the

classification standard category. ................................................................................................... 98

Figure 3.16. The Spectral Information Divergence (SID) map shows fair differentiation between

rock units. The accuracy assessment of the SID shows an overall accuracy with 52%; the Kappa

Coefficient is 0.47, which is known as “fair results" in the category. .......................................... 99

Figure 3.17. The Spectral Angle Mapper (SAM) classification performs the second best result

with an Overall accuracy of 61%. The calculated Kappa Coefficient is 0.56 placing into the good

agreement category of the Kappa Coefficient statistic. .............................................................. 100

Figure 3.18. The Support Vector Machine (SVM) digital-mapping shows the best result with an

overall accuracy of 86.4%. The calculated Kappa Coefficient is 0.85 placing into the excellent

agreement category of the Kappa Coefficient statistic. .............................................................. 101

Figure 3.19. Geological map of Al-Uyaynah area showing the mapped area of extrusive

carbonatites (modified from [1]. ................................................................................................. 102

Figure 3.20. A comparing of confusion matrix over all accuracy, and kappa coefficient results.

The binary encoding classification (BEC) overall accuracy is 43.59%; the calculated Kappa

Coefficient is 0.32. This recognized as No agreement ‘not even poor results’ in the classification

standard category. The Spectral Information Divergence (SID) map shows a good

discrimination among ferro/calcio-carbonatite and other the rock units. The accuracy assessment

of the overall accuracy is 68.3%; the Kappa Coefficient is 0.58, which is known as "good

results" in the category. SAM performs the second best result with an Overall accuracy of

81.54%. The calculated Kappa Coefficient is a mere 0.75 placing into the very good agreement

category of the Kappa Coefficient statistic. The Support Vector Machine (SVM) digital-mapping

results with an overall accuracy of 97.58%. The calculated Kappa Coefficient is 0.97. This

recognized as excellent in the classification standard category. ................................................. 103

Figure 3.21. A Comparison between the rocks unit’s distribution that classified correctly, in

percent, of BEC map based on the results of the confusion matrix accuracy assessment. BEC

method gave, overall, the poorest result after comparing BEC to the field ground truth regions.

..................................................................................................................................................... 114


percent, of SID map based on the results of the confusion matrix accuracy assessment. SID

method gave, overall, a fair result after comparing SID to the field ground truth regions. ........ 115

xiv

Figure 3.23. A Comparison between the rocks units’ distribution that classified correctly, in

percent, of SAM map based on the results of the confusion matrix accuracy assessment. SAM

method performs the second best result, overall good agreement, after comparing SAM to the

field ground truth regions............................................................................................................ 116


percent, of SVM map based on the results of the confusion matrix accuracy assessment. SVM

method performs the second best result, overall an excellent agreement, after comparing SVM to

the field ground truth regions. ..................................................................................................... 117

Figure 3.25. Shows the ferro-carbonatite layers that intruded into metasediments unit of the Hatta

Zone. Modified from [3] ............................................................................................................. 118

Figure 3.26. shows the location and the geological map of the Hatta zone Modified from [2, 3]

..................................................................................................................................................... 119

Figure 3.27. The Support Vector Machine (SVM) digital-mapping of the Hatta carbonatite,

shows the best result with an overall accuracy of 99.85%. The calculated Kappa Coefficient is

0.99.............................................................................................................................................. 120

Figure 4.1. The range of the electromagnetic spectrum, showing most crucial ranges for

minerals sensing spectroscopy applications. The visible/ near infrared (VNIR), and short-wave

infrared (SWIR), modified from [17]. ........................................................................................ 132

Figure 4.2. A field view of Al-Uyaynah ferro-carbonatites, (a) shows the ferro-carbonatites


(d) hand sample from ferro-carbonatite area. ............................................................................. 135

Figure 4.3. A field pictures of Al-Uyaynah calcio-carbonatites (a) shows the calcio-carbonatites

outcrops of the northern part of the study area, (b, and c) show a close view of clacio-

carbonatites. (d) a hand sample calcio-carbonatite. .................................................................... 136

Figure 4.4. Example of the field spectra of ferro-carbonatite. The original reflectance graph in

red, and the reflectance after applying the continuum removal method in green. Absorption

depth, area, and center are labelled. ............................................................................................ 139

Figure 4.5. Shows calcite absorption bands in USGS spectra library (left) [36] , and calcite

absorption area which recorded in calcio-carbonatite (right). The absorption was observed in the

area between 2.312–2.345 μm, and the deepest absorption point was recorded at 2.34 μm of fresh

calcio-carbonatite. ....................................................................................................................... 142

Figure 4.6. Illustrates calcite absorption bands in ferro-carbonatites. The absorption was between

2.312–2.345 μm of the spectra wavelength, and the deepest absorption point was at 2.34 μm of

much of ferro-carbonatites. ......................................................................................................... 143

xv

Figure 4.7. Shows iron oxides from USGS spectra library (left) [36] , and the iron oxide

absorption area that is observed in ferro-carbonatite (middle). The iron absorption was recorded

in the area between 0.8-1.26μm. The deepest absorption point was recorded around 0.885 μm of

fresh ferro-carbonatite and 1.037 μm for the weathered part. The calcio-carbonatite shows

shallow absorption in both fresh and weathered surfaces, however, the iron oxide absorption was

more pronounced in weathered outcrop. ..................................................................................... 146

Figure 4.8. Shows neodymium (Nd) from USGS spectra library (left) [36], and the Nd

absorption area that was observed in ferro-carbonatite (right). The Nd absorption spanned the

area from 0.7-0.78μm. The deepest absorption was recorded around 0.74 μm. ......................... 147

Figure 4.9. Neodymium (Nd) absorption area that was observed in ferro-silico-carbonatite

weathered surface (left) and fresh interior (right). The Nd absorption depth is higher (at 0.705

µm) in fresh surfaces in comparison to weathered surfaces. ...................................................... 148

Figure 4.10. Nd absorption in calcio-carbonatite was verified. The deepest absorption point was

recorded around 0.752μm of the weathered surface of the calcio-carbonatite spectrum. The un-

weathered part shows a shallower absorption at 0.751 μm......................................................... 149

Figure 4.11. Illustrates The CaO, Fe2O3 and SiO2 concentration in the weathered exterior and

fresh interior from selected field samples of Al-Uyaynah carbonatite. The CaO is higher in

calcio-carbonatite, while the iron oxides are greater in ferro-carbonatites. The weathered part of

carbonatites contains more iron oxides than the fresh part of the rock. SiO2 has higher

concentrations in ferro-carbonatite samples versus calcio-carbonatites. .................................... 152

Figure 4.12. REE distribution patterns of carbonatite from Al-Uyaynah (this study) compared

with those previously published (Woolly. 1991)., and the global average of carbonatite. The

values are normalized to C1-chondrite [40]. The diagram (a) illustrates the REE distribution of

ferro-carbonatite that were collected from fields vs the Al-Uyaynah carbonatites of Woolley

1991 [39], and world average of ferro-carbonatite. (b) shows the REE distribution of calcio-

carbonatite and average global and Woolley average of calcio-carbonatites. ............................ 161

Figure 4.13. Calcite absorption center ranges measured from the carbonatite spectra. ............. 164

Figure 4.14. Statistical comparison between calcite concentrations of carbonatites and the

equivalent laboratory spectra absorption depth records. The change in absorption depth due to

the increase the calcite concentrations of carbonatite samples spectra measured with the ASD in

the laboratory. The R2 ~ 0.8 (slope close to 1) was found among measured depth and chemically

measured calcite concentrations in weight percentage. .............................................................. 164

Figure 4.15. Displays the absorption center ranges of neodymium measured from the carbonatite

spectra. ........................................................................................................................................ 165

Figure 4.16. Relationship between neodymium concentrations (ppm) and the spectra absorption

depth of carbonatites spectra. The neodymium absorption depths were calculated based on

xvi

continuum removal method of carbonatite spectra, and Nd concentrations were measured using,

LA-ICP-MS. The coefficient of determination of the Nd concentration vs its absorption depth

displays a confident correlation (R2 > 0.9). ................................................................................ 165

Figure 4.17. Shows the absorption center ranges of iron oxides measured from the carbonatite

spectra. ........................................................................................................................................ 166

Figure 4.18. The iron oxides absorption feature depths of the carbonatite spectra were evaluated.

The regression relationships were used to understand the connection between iron oxides

absorption depths and its concentrations in carbonatites. The change in absorption depth with the

increase of the iron concentrations in carbonatites has good correlation (R2 > 0.8). ................. 166

Figure 4.19. Shows the absorption center ranges of Nd measured from the carbonatite spectra.

..................................................................................................................................................... 167

Figure 4.20. Nd concentration (in ppm) plotted against the spectra absorption depth of

carbonatites from this study and others [29]. The neodymium absorption depths were calculated

based on continuum removal method of carbonatite spectra. The coefficient of determination of

Nd concentration vs its absorption depth shows a positive correlation (R2 > 0.6). .................... 167

Figure 4.21. Shows Nd and iron concentration (in ppm) plotted against spectra absorption depth.

In the samples with high REE-Nd concentration and low iron, the absorption is greater in depth.

..................................................................................................................................................... 168

1

1. CHAPTER 1

DISSERTATION PHASES AND BACKGROUND

1. Introduction

Remote sensing science is characterized by the evaluation and examination of data

obtained from indirect physical interaction with the materials or objects [13]. In this study, remote

sensing denotes the procedures used to measure electromagnetic energy as photons reflected by

earth surface materials. Remote sensing acquires electromagnetic energy in distinct ranges of

wavelength bands [14]. Remote sensing sensors usually analyze the electromagnetic energy at

distinct wavelengths that are reflected, scattered, and emitted from Earth’s materials all through

the electromagnetic energy spectrum, from nearly 0.3 to 15 μm[15]. The reflective spectrum covers

the ranges of about 0.38 to 3.0 μm; it characterizes that portition of the solar reflected energy, or

spectrum, that is utilized specifically in remote sensing [15], and especially useful in identifying

geological materials and mapping. According to many scholars e.g., [14, 16], differences in the

electromagnetic spectrum are likely the consequence of the interaction between the

electromagnetic waves and components of the rocks/materials (e.g., elements and minerals). This

process thus allows for the remote recognition of Earth materials by detecting the reflected energy,

and also possibility of mapping of the distribution of these minerals based on their spectral

characteristics. The electromagnetic energy emitted or radiated from the sun has been modeled by

2

a 6000◦ K blackbody curve [14]. Solar radiation or energy passes through the atmosphere in

particular transmission windows. These windows allow the shorter visible (VIS) from 0.38 – 0.72

μm, near-infrared (NIR) from 0.72 – 1.3 μm, and shortwave-infrared (SWIR) from 1.4 – 2.5 μm,

to reach the earth surface, which are then partially absorbed and re-emitted or reflected. It is theese

reflected waves that are recorded by remote sensing sensors [14, 15].

Since the early 1960s, the scope of remote sensing has broadened extraordinarily in terms

of its application to geological mapping and mineral exploration. At its inception, remote sensing

was employed as a quick and cost-effective way to analyze to spatial variations in lithology [17,

18]. During the 1970s, geologists employed remote sensing techniques using airborne systems

[19]. Further advances came in the early 1980s with the development of hyper-spectral methods

for Earth analysis [20]. The spectral resolution used in the modern-day remote sensing has greatly

improved and is now particularly useful for distinguishing mineral types and rock groups [16, 18,

20]. Specifically, it has been shown that a broad range of minerals such as carbonates, clays, and

hydroxides can be sensed through this method (e.g. [21]). The empirical literature showed the

capability of the remote sensing approaches to identify and map minerals and rocks based on its

particular spectra which are significant for exploring the mineral deposits [22-24]. The

multispectral platforms employed in the past, i.e., TM and Landsat MSS, have been employed for

analyzing structures and common geomorphic surveys on a large scale [24, 25]. The Landsat TM

system has been employed for the surveying of minerals by using the absorption features in two

SWIR bands sensitive to alteration minerals in particular [25, 26]. The Advanced Spaceborn

Thermal Emission and Reflection Radiometer (ASTER) expanded remote sensing capabalities to

include the visible/near infrared (VNIR), short-wave infrared (SWIR), and thermal infrared (TIR)

3

spectra, which have great significance in mineral surveying [27]. The ASTER sensor is capable of

detecting more bands than previous sensors, which allows geologists to chart rocks and minerals

on the surface of the earth with more precision and accuracy [26, 28].

The ranges between 0.45 to 2.5 µm of the reflectance spectra wavelengths are considered

an essential part of the spectrum for geo-remote sensing applications (Figure 1.1). These ranges

can be utilized in defining the composition of minerals on the basis of absorption characteristics

[29-31]. Carbonatites that are enriched with about fifty percent or more of calcite minerals display

a strong absorption feature of the CO32- ion, which is recognized easily within the range of 2.2-

2.33 µm [32]. The iron minerals and REE elements, which may found in ferro-carbonatite

in abundance, also impact reflectance spectra. The iron absorption characteristics exhibit a

distinctive absorption and are pronounced within the window of 0.9-1.3 µm [21]. REE absorption

features are principally linked to electronic transitions in Nd3+, which generates absorption

characteristics in the wavelength ranges between 0.5 and 0.8 um [30, 33]. A blend of these

absorption features could yield a basis for the creation of carbonatite maps using remote-sensing

data.

The reflection spectra properties of carbonatites have previously been applied to prior

investigations to produce maps of carbonatite complexes using aircraft sensors. AVIRIS (Airborne

Visible / Infra-Red Imaging Spectrometer) data were employed to identify the Ice River

carbonatite complex and showed slight absorption properties

4

Figure 1.1. Illustration of the ranges of the electromagnetic spectrum showing the most crucial ranges for

mineral sensing spectroscopy applications - the visible/near infrared (VNIR), and short-wave infrared

(SWIR) [10].

5

located at 0.58, 0.74, 0.80, and 0.87 um, which suggest a neodymium absorption feature [34].

Likewise, novel investigation of the Sarfartoq carbonatite in Greenland was carried out

by adopting a mix of spectral techniques to separate between carbonatite, fenite, and hematised

rocks, with a special focus on the Nd3+ reflectance feature [35]. Generally, little attention has been

paid to detecting carbonatite occurrences using satellite-based sensors.

In this research, we have focused on examining in what way remote sensing approaches can be

employed to observe carbonatite distribution. These rocks, which contain minerals rich in rare-

earth elements (REE), have substantial economic significance. This study analyzed the spectral

characteristics of carbonatites by employing Landsat 8 and ASTER datasets to map the spatial

distribution of Al-Uyaynah carbonatites in UAE. This research offers new ways to more efficiently

understand carbonatite spectra and map carbonatite distribution. We anticipate that our work will

result in the development of techniques for REE ore deposit exploration and investigation.

2. Carbonatite

2.1. Carbonatite definition and subtypes

Carbonatite is defined based on its major minerals, and in the IUGS system of

categorization as igneous rock that comprised of carbonate minerals with modal percent exceeding

50 % [36]. The sub-types of carbonatite are defined as below:

6

(i) A rock is referred to as a calcite carbonatite, or calcio-carbonatite, when the dominant

carbonate mineral is calcite. The term sövite is applied if the calcio-carbonatite is coarse-grained,

and alvikite if it is fine-grained. Although both sövite and alvikite are calcio-carbonatites, they can

exhibit distinctly different chemical compositions with regards to trace elements [37].

(ii) The “ferro-carbonatite” name was used for around thirty years, however, is not defined

adequately [38]. The IUGS system determines it as a carbonatite mineralogically where the major

carbonate minerals are rich in iron, but chemically was defined as a carbonatite [36, 38]. Gittins

and Harmer recognized these difficulties and recommended the use of the term “ferro-carbonatite”

for Fe-rich carbonatites only for cases in which the modal mineralogy of the rocks is unknown i.e.

that the ferro-carbonatite expression should be utilized, and use of terms such as “siderite

carbonatite, magnetite-calcite carbonatite, etc. for cases in which the modal mineralogy is known

[38].

(iii) Magnesium-rich carbonatites are dolomite-bearing and occasionally are referred to as

“beforsite,” although this term is becoming obsolete; the terms dolomite carbonatite or magnesio-

carbonatite are currently preferred [39].

(iv) Natro-carbonatite is a fine-grained carbonatite lava composed primarily of nyerereite and

gregoryite [(Na, K)2CO3]. This is an exceptionally rare carbonatite form, and its occurrence is at

present restricted to Oldoinyo Lengai, Tanzania [40].

7

Figure 1.2. Global distribution of Nb– and REE–bearing carbonatite deposits [6].

Figure 1.3. Classification diagram of carbonatite based on its chemistry; modified from [12].

8

Carbonatites have been found at many locations worldwide. These locations were plotted

on a global map [6] by the U.S. Geological Survey (USGS) Mineral Resources Program (Figure

1.2). In this research our study area, Al-Uyaynah, is one of these areas where carbonatite is known

to exist. It was discovered by Woolley and others in 1991. They found two particular magma types

(calcio and ferro-carbonatites), that comprise the Al-Uyaynah carbonatite [41]. These two varieties

of carbonatite, represent the majority of carbonatites found globally (Figure 1.5, and 1.7), and

therefore they are useful analogs for studies of carbonatites more broadly. We utilize existing

geological studies of our region to provide structural, mineralogical, and chemical data that are the

basis for the current research effort [41].

2.2 Carbonatite petrogenesis

The concept of carbonatite magma was initially proposed by Eckermann in 1948 [42, 43].

It was not, however, fully accepted by the scientific community at that time [42, 43]. In 1960-61,

sodium-carbonate lava erupted from the Oldoinyo Lengai volcano in Tanzania[44]. The

observation and successful experimental studies performed on carbonatite-melts (e.g. [45-47]),

which seemed to confirm their existence, settled many of these initial arguments. Globally in 1987

about 56 carbonatite occurrences had been identified [48, 49], but this number has grown and there

are currently more than 500 carbonatites known [50]. Since carbonatite was initially discovered,

many arguments have centered on the existence and definition of carbonatite magmagenesis [51-

53], and these issues persist to this day [54]. Despite the continued ambiguity in terms of

carbonatite origin, studies on these unusual rocks have progressed dramatically in terms of our

9

undertanding of their phase stability, major and trace element characteristics and isotopic

fingerprints [55].

In the period before 1960, there was little recognition of the role of the mantle in the

genesis of carbonatites and indeed the role of the mantle for the genesis of very alkalic rocks in

general [54]. In the early 1960s, the earth’s mantle was documented as the source of the majority

of carbonatite magmas, as the newly collected 87Sr/86Sr data on carbonatites showed that it was

actually of mantle origin [54]. While the utimate origin of carbonatites is considered to be mantle-

derived, the precise mechansims by which carbonatite magmas are generated is still uncertain.

Most modern scholars favor three primary carbonatite generation mechanisms: 1) primary melt

product of mantle; or 2) derivation by liquid-immiscibility (immiscible separation of carbonate

liquid from an alkalic magmas); 3) the final-products of low-pressure silicate-melt fractional

crystallization [52]. In our study area, Al-Uyaynah carbonatite petrochemical data shows a mantle

origin for the Al-Uyaynah carbonatites, which has been previously interpreted to have been formed

in a volcanic island environment set within a transitional region of continental to oceanic crust [11,

41, 56].

2.3 Rocks spatially connected with carbonatites

Previous authors have defined the common rock types that are usually found in and around

a carbonatite deposit. The majority of rocks that are spatially associated with carbonatites are

generally metasomatic and magmatic rocks. From the over 350 reported localities, carbonatites are

usually associated with the silicate rocks of quite variable composition [57, 58]. The common rock

10

Figure 1.4. Displays the tectonic setting of Al-Uyaynah area within Arabian plat. (modified from [7]).

11

types are divided into five groups: 1) alkaline rocks, 2) alkaline ultramafic and mafic rocks,

3) alkaline nepheline-feldspathoid syenitic rocks, 4) metasomatic rocks, and 5) carbonatitic rock

sedimentary deposits that are associated with the carbonatite itself (Figure 1.8 and Table. 1) [58].

2.4 Carbonatite tectonic settings

Carbonatite deposits are ordinarily restricted to a stable continental portion of a tectonic

plate, i.e. among the well-developed old crust of the Earth such as cratons, shields, and crystalline

blocks within cratons [59]. Carbonatites are commonly restricted to alkaline magmatic provinces

controlled by intracontinental rift and fault systems within these broad regional structures [6, 58,

59]. Some carbonatite deposits are exceptions and can occur in a Foreland Belt through the western

edge of the North American continent [6, 60]. Also there are two identified other carbonate

occurrences (particularly relevant to this study) that are related to the oceanic basins (Canary

Island and Cape Verde Islands) [59].

Generally, the structural framework of our focus site - the Al-Uyaynah zone - results from a

sequence of thrust sheets, dipping to the east and southeast [56]. These Mesozoic thrust sheets

were formed at some point during the closing of the southern Tethys Sea. The latest formed

oceanic-crust, which is characterized now as the Semail Ophiolite, was obducted over the north-

eastern portion of the margin of the Arabian plate (Figure 1.4) [41, 61, 62]. The sediments and

volcanic rocks between the Arabian continent and the oceanic crust slab were amalgamated as

narrow thrust layers within the accumulated allochthonous unit [41, 61, 62]. These young layers

were gradually thrust and folded toward the south-west direction [41]. As a result, the Al-Uyaynah

12

Figure 1.5. Map of the northern eastern of UAE indicating the location of the Al-Uyaynah area and general geological units left, modified from [1, 2]. (b)

ASTER 3,2,1 natural color image right.

13

zone carbonatite and the associated metamorphic rocks were structurally covered by the Semail

Nappe, and were then later thrust over the continental margin sediments [11, 41, 56, 61].

2.5 Economic significance of carbonatite

Major mineral deposits associated with carbonatite rocks are counted as one of the ultimate

rare earth elements sources. Rare earth elements (REEs) are composed of the lanthanide elements

plus scandium and yttrium, which have comparable physical characteristics and are often observed

in the same deposits [59]. REEs comprise the light REE (LREE) elements from lanthanum to

gadolinium, and heavy REEs (HREEs) terbium to lutetium [58, 59]. The most abundant source of

the Light REE from Lanthanum (La) to Europium (Eu), are in difficult to find REEs bearing

minerals like monazite-Cerium (Ce) and bastnasite/bastna¨site (Ce) (Table. 2), which produce

about 70- to 80% of the global LREE supply [9]. The heavy REEs from Gadolinium (Gd) to

Lutetium (Lu), are frequently found in accessory minerals in alkaline and acidic rocks [9, 63].

REEs are not uncommon; however, economically extractable deposits are rare [58].

During past few decades, the needs for REEs have expanded significantly because of their

extensive and varied uses in leading technology products, such as green energy products. The

importance of these deposits has resulted in a growing economic rivalry between suppliers and

consumers of REE, and also raised environmental concerns due to the increase in mining activities.

For much of the 20th century, the Mountain Pass deposit, in the United States, was the world's

predominant source of REEs [64]. However, with declining activities there, the U.S. has turned to

be imports of REEs from another countries (Brazil 37%, China 25% of global REEs deposits) [65].

14

Global REE production comes from just a few locations (Figure 1.2). As a consequence, exploring

and locating new sources of REE deposits has become a critical priority [58].

Igneous rocks, like carbonatites, are recognized as important hosts of REEs, which

manifest in an array of unusual minerals [12, 66]. The REEs are found in carbonatite accessory

minerals, which include biotite, ankerite, amphiboles, hematite, pyroxenes, ilmenite, pyrite, etc.

[58]. The REEs are typically concentrated in calcio, mangiesio and ferro-carbonatites, however,

the highest REEs abundances are commonly found in ferro-carbonatite [12]. REEs are typically

concentrated in carbonatite REEs bearing minerals. The highest percentage of REEs are founded

in bastnäsite 53–79 wt%, parasite 58–63 wt%, and synchysite 48–52 wt% [9].

In our study area, calcite is the major mineral phase in the calcio-carbonatites, which also

include a significant amount of apatite, as well as opaque minerals such as magnetite. The typical

composition of the calcio-carbonatites lapilli is 40% calcite, 20-30% apatite, and 20-40% opaque

minerals [41]. The Al-Uyaynah ferro-carbonatites are richer in silicate, iron minerals, and (REEs)

compared to calcio-carbonatites. The ferro-carbonatite comprises of calcite, amphibole, biotite,

opaque minerals, and abundant allanite. The allanite in ferro-carbonatites usually occurs as crystal

accumulations, and the REEs are concentrated in these accumulations as allanite-Ce [41].

15

Figure 1.6. Rare elements (REEs) and selected high field strength element distribution in the mantle and

earth’s crust, which is normalized to CI chondrite [9].

16

Figure 1.7. Geological map of Al-Uyaynah area showing the extrusive carbonatites (modified from [8]).

17

3. Detecting, assessing, and analyzing carbonatites and surrounding rock units at Al-

Uyaynah, UAE using remote sensing techniques.

The initial aim of this investigation was to examine the remote sensing techniques for

detecting and mapping carbonatites distribution in the in Al-Uyaynah region at UAE. The research

surveys carbonatites, by employing the Visible/Near Infra-Red (VNIR) and Short Wave Infra-Red

(SWIR), with two datasets: (1) the latest multispectral sensor (Landsat 8 OLI), and (2) ASTER

imaging.

The first study carries out an examination of Landsat 8 OLI data to map the distribution of

carbonatites. This approach is novel given that the Al-Uyaynah area had not been explored prior

to this study with Landsat 8 imagery through mapping algorithms. The SAM algorithm was

employed to locate and to identify carbonatite occurrences and map carbonatite distribution. Our

results were compared with available geological maps and field ground truth data to evaluate the

utility of Landsat 8 imagery ability in carbonatite mapping.

The second study examines ASTER imaging to also map carbonatites in Al-Uyaynah area.

In this experiment, four different data processing techniques were used to identify carbonatites:

Spectral Information Divergence (SID), Support Vector Machine (SVM), Spectral Angle Mapper

(SAM), and the Binary Encoding Classification (BEC). These algorithms were used, to classify

the study area using ASTER image, digitally map carbonatites and adjust rock components. The

results are illustrated in digitally-produced maps which highlight the pixels that correspond to

specific endmember-spectra among other classes of a different kind. Out of the methods tested,

18

Figure 1.8. A comparison between the occurrence of rocks accompanying carbonatites [Count is number of deposits; % is percentage of deposits with related

rock type, data used in this diagram were obtained from table3 of USGS Open-File Report [6]

19

Rock group Rocks Count %

Carbonatitic Carbonatitic breccia 18 30

Carbonatitic agglomerate 7 12

Alkaline Syenite 20 33

Trachyte 9 15

Nordmarkite 4 7

Alkaline granite, pegmatite 3 5

Alkaline nepheline-feldspathoid

syenitic

Nepheline-syenite 21 35

Ijolite 16 27

Melteigite 7 12

Phonolite 5 8

Foyaite 4 7

Nephelinite 3 5

Urtite 3 5

Malignite 2 3

Shonkinite 2 3

Okaite 2 3

Tephrite 1 2

Melilitolite 1 2

Alkaline ultramafic and mafic Pyroxenite (incl. bebedourite) 17 28

Lamprophyre, monchiquite, alnoite 9 15

Phoscorite 7 12

Jacupirangite 6 10

Gabbro, diabase 6 10

Peridotite 5 9

Tuff 5 9

Dunite 4 7

Basalt 1 2

Metasomatic Fenite 7 12

Glimmerite 5 8

Albitite 3 5

Silexite 1 2

Table 1.1. Occurrence of rocks accompanying carbonatites [Count is number of deposits; % is percentage of

deposits with related rock type, data were acquired from USGS Open-File Report [1].

20

Mineralb Formulac Rare element

(wt% range or

max. content)

Major deposit type(s)d Localities: key examples (past, present, and

potential producers)

Bastna¨site LREECO3(F,OH) 53–79 SREE2O3 Carbonatites and associate metasomatic

rocks, altered peralkaline feldspathoid

rocks

Mountain PassU, Bayan OboCh, WeishanCh,

MaoniupingCh,

NechalachoCa

Parisite CaLREE2(CO3)3(F,OH)2 58–63 SREE2O3 Carbonatites and associate metasomatic

rocks, hydrothermal deposits

Mountain PassU, Bayan OboCh, WeishanCh,

SnowbirdU

Synchysite CaREE(CO3)2(F,OH) 48–52 SREE2O3 Carbonatites and associate metasomatic

rocks, altered peralkaline feldspathoid

and granites

Barra do Itapirapua˜B, Lugiin GolM, Ak-TyuzK,

NechalachoCa

Monazite (LREE,Th,Ca)(P,Si)O4 38–71 wt%

SREE2O3

Carbonatites and associate metasomatic

rocks

Mountain PassU, Bayan OboCh, EneabbaA, Mt

P-rich nelsonite, weathering crusts;

placers

Mount Weld and WIM 150A, KangankundeMa,

TomtorR,

SteenkampskraalSA, ManavalakurichiI

Xenotime (HREE,Zr,U)(P,Si)O4 43–65 SREE2O3 Carbonatites and associate metasomatic

rocks, weathering crusts, placers

LofdalN, Ak-TyuzK, PitingaB, TomtorR, Mt Weld

and WIM 150A, Kinta and SelangorMs

Churchite

Gadolinite

HREEPO42H2O

REE2FeBe2Si2O10

43–56 SREE2O3

45–54 SREE2O3

Weathering crusts

Granitic pegmatites

ChuktukonR, Mt WeldA

YtterbyS, Strange LakeCa, Barringer HillU

Rutile

Loparite

Fergusonite

Columbite–

tantalite

(Ti,Nb,Ta,Fe,Sn)O2

(Na,REE,Ca,Sr,Th)

(Ti,Nb,Ta)O3

REENbO4

(Fe,Mn,Mg)(Nb,Ta,Ti)2O6

56 Ta2O5, 34

Nb2O5,

7 SnO2

28–38 SREE2O3,

20

Nb2O5, 1 Ta2O5

43–57 SREE2O3,

40–55

Nb2O5, 0.8 Ta2O5

Carbonate metasomatic rocks, granitic

pegmatites, placers, weathering crusts

Peralkaline feldspathoidal rocks

Metasomatic carbonate and peralkaline

feldspathoid rocks, granitic pegmatites

Carbonatites and associate metasomatic

rocks, granites, and granitic pegmatites,

placers

Bayan OboCh, GreenbushesA, Kinta ValleyMs,

Morro dos Seis Lagos, and BorboremaB

Karnasurt and UmbozeroR

Bayan OboCh, Barringer HillU, NechalachoC

Blue RiverCa, Bayan OboCh, Greenbushes and

WodginaA, Koktokay and YichunCh, Pitinga and

MibraB, KentichaE, MarropinoMz, Nord-Kivu

and Sud-KivuDRC

Mineralb Formulac Rare element

(wt% range or

max. content)

Major deposit type(s)d Localities: key examples (past, present, and

potential producers)

Tapiolite

Wodginite

(Fe,Mn)(Ta,Nb)2O6

(Mn,Fe)(Sn,Ti)(Ta,Nb)2O8

72–86 Ta2O5, 9

Nb2O5 56–85

Ta2O5, 15

Granitic pegmatites

Granitic pegmatites

TancoCa, GreenbushesA

TancoCa, Greenbushes, and WodginaA

Table 1.2. Shows the rare earth element (REE)- Major bearing minerals [9]

21

Table 1.2. (cont’d)

Nb2O5, 3–18 SnO2

Ixiolite (Ta,Nb,Mn,Fe,Sn,Ti)4O8 70 Ta2O5, 72 Nb2O5,

20 SnO2

Granitic pegmatites TancoCa, BorboremaB

Pyrochlore (Ca,Na,Sr,Ba,Pb,K,U)2x

(Nb,Ti,Ta,Zr,Fe)2O6

(F,OH)1y nH2O

29–77 Nb2O5, 16

Ta2O5, 22 wt%

REE2O3

Carbonatites and associated phoscorites

Peralkaline granites and associated

Pegmatites, fenites, weathering crusts

Barreiro and Catalaõ I and IIB, Oka, Niobec

and

Strange LakeCa, Tomtor, Chuktukon,

Tatarskoye, Bol’shetagninskoye and Belaya

ZimaR, Lueshe and Nord-KivuDRC, PitingaB

Microlite

Baddeleyite

(Ca,Na,Pb,U,Sb,Bi)2x

(Nb,Ta,Ti)2O6(OH,F)1y

(Zr,Hf,Nb,Fe)O2

46–81 Ta2O5, 20

Nb2O5, 9 SnO2

88–99 ZrO2, 4.8

HfO2,

6.5 Nb2O5

Granites and granitic pegmatites

Phoscorites, altered peralkaline

feldspathoid syenites, carbonate

metasomatic rocks, placers

TancoCa, GreenbushesA, Koktokay and

YichunCh

Kovdor and AlgamaR, PalaboraSA, Poços de

CaldasB

Zircon (Zr,Hf,HREE,Th,U)

(Si,P)O4

64–67 ZrO2, 1.5

HfO2,

19 SREE2O3

Placers; peralkaline, feldspathoid

syenites

(including altered varieties)

Jacinth-Ambrosia and EneabbaA, Richards

BaySA, Manavalakurichi and ChavaraI, Poços

de CaldasB, NechalachoCa

22

the high accuracy rates (successful algorithms) were defined as the ones to offer the best ways

for carbonatite mapping.

4. Understanding carbonatite chemistry using spectral analysis devices (ASD) and

geochemical approaches.

The second goal of the investigation was to increase our understanding of VNIR and SWIR

spectral frameworks of carbonatite rocks through petrochemical analysis, to build a more

comprehensive understanding of chemical environments of the Al-Uyaynah carbonatite, UAE.

Spectroscopy is a quick and affordable way of analyzing the carbonatite mineralogy. In this

research, the carbonatite samples were evaluated and examined by utilizing the Analytical Spectral

Device (ASD) device. Inductively Coupled Plasma Mass Spectrometry (ICP-MS) procedures were

used to determine rare-earth elements (REE) and other trace elements of the whole rock samples.

X-ray Fluorescence (XRF) was utilized to investigate the whole-rock composition major oxides.

These geochemical analyses permit us to test linearly the relationship between the major

oxides/elements absorption features of carbonatite spectra and its concentration in carbonatites.

5. Dissertation phases

The research is written within a manuscript format with a view to enable the current

researcher to compile and submit individual research papers derived from its chapters to peer-

reviewed journals. The current effort focuses initially on the utilization of SWIR and VNIR of the

23

multispectral image data to map carbonatites and neighboring lithologies in Al-Uyaynah area,

UAE (Chapter 2, 3).

In the final phase of this study, ASD reflectance spectra data were obtained through hand-

sample from the carbonatites in of Al-Uyaynah, UAE. These samples were analyzed and compared

to find out the ideal wavelength ranges that can assist us in detecting, identifying, and mapping the

ferro/calcio-carbonatites. The research was aided by a geochemical analysis by employing XRF,

ICP-MS to offer a comparative perspective to ASD petrogenetic properties and to assess the

carbonatites spectra (Chapter 4). The final chapter, Chapter 5, provides a synthesis of the entire

research results as a whole and suggestions for future research area suggested in these two fields.

24

REFERENCES

25

REFERENCES

[1] Ryan, W.B., et al., Global Multi‐Resolution Topography synthesis. Geochemistry,

Geophysics, Geosystems, 2009. 10(3).

[2] Tarapoanca, M., et al., Forward kinematic modelling of a regional transect in the Northern

Emirates using geological and apatite fission track age constraints on paleo-burial history.

Arabian Journal of Geosciences, 2010. 3(4): p. 395-411.

[3] Berger, V.I., D.A. Singer, and G.J. Orris, Carbonatites of the World, Explored Deposits of

Nb and REE--database and Grade and Tonnage Models. 2009: US Geological Survey.

[4] Stern, R.J. and P. Johnson, Continental lithosphere of the Arabian Plate: a geologic,

petrologic, and geophysical synthesis. Earth-Science Reviews, 2010. 101(1): p. 29-67.

[5] Goodenough, K.M., et al., Geology of the Khor Fakkan 1: 50 000 map sheet 50-3, United

Arab Emirates. 2006: United Arab Emirates, Ministry of Energy, Petroleum and Minerals

Sector, Minerals Department.

[6] Linnen, R., et al., Geochemistry of the rare-earth element, Nb, Ta, Hf, and Zr deposits.

Treatise on geochemistry, 2014. 13: p. 543-568.

[7] Kerr, A., et al., Visible/infrared spectroscopy (VIRS) as a research tool in economic

geology: background and pilot studies from NewFoundland and Labrador. Geological

Survey, Report, 2011. 11: p. 145-166.

[8] Woolley, A. and D. Kempe, Carbonatites: nomenclature, average chemical compositions,

and element distribution. Carbonatites: genesis and evolution. Unwin Hyman, London,

1989. 619.

[9] Jensen, J.R., Introductory digital image processing: a remote sensing perspective. Vol.

3rd. 2005, Upper Saddle River, N.J: Prentice Hall.

[10] Elachi, C. and J.J. Van Zyl, Introduction to the physics and techniques of remote sensing.

Vol. 28. 2006: John Wiley & Sons.

[11] Campbell, J.B., Introduction to remote sensing. 2002: CRC Press.

[12] Clark, R., Spectroscopy and principles of spectroscopy, manual of remote sensing. John

Wiley and Sons, Inc, 1999.

26

[13] Bauer, M.E., et al., Spectral-agronomic relationships of corn, soybean and wheat canopies.

1981, West Lafayette, Ind. Springfield, Va.: Purdue University for sale by National

Technical Information Service. 17 p.

[14] Hunt, G.R., and J. W. Salisbury Visible and near-infrared spectra of minerals and rocks.

1971.

[15] Williams Jr, R. The use of broadband thermal infrared images to monitor and to study

dynamic geological phenomena. in Geological Applications of Thermal Infrared Remote

Sensing Techniques. 1981.

[16] Goetz, A.F.H., Imaging spectrometry for earth remote sensing. Science, 1985. 228(4704):

p. 1147.

[17] Van der Meer, F., Spectral reflectance of carbonate mineral mixtures and bidirectional

reflectance theory: Quantitative analysis techniques for application in remote sensing.

Remote Sensing Reviews, 1995. 13(1-2): p. 67-94.

[18] Kruse, F.A., et al., The spectral image processing system (SIPS)—interactive visualization

and analysis of imaging spectrometer data. Remote Sensing of Environment, 1993. 44(2-

3): p. 145-163.

[19] Kruse, F.A., J.W. Boardman, and J.F. Huntington, Comparison of airborne hyperspectral

data and EO-1 Hyperion for mineral mapping. Geoscience and Remote Sensing, IEEE

Transactions on, 2003. 41(6): p. 1388-1400.

[20] Kaufmann, H. Mineral exploration along the Aqaba-Levant structure by use of TM data;

concepts, processing and results. 1988. United Kingdom (GBR): Taylor & Francis,

London, United Kingdom (GBR).

[21] Abdelsalam, M.G., R.J. Stern, and W.G. Berhane, Mapping gossans in arid regions with

Landsat TM and SIR-C images: the Beddaho Alteration Zone in northern Eritrea. Journal

of African Earth Sciences, 2000. 30(4): p. 903-916.

[22] Zhang, X., M. Pazner, and N. Duke, Lithologic and mineral information extraction for gold

exploration using ASTER data in the south Chocolate Mountains (California). ISPRS

Journal of Photogrammetry and Remote Sensing, 2007. 62(4): p. 271-282.

[23] Abrams and Hook, ASTER User Handbook,http://asterweb.jpl.nasa.gov/documents.asp

(accessed 21 January 2008) 2002, NASA.

[24] Meer, F.v.d. and S.M.d. Jong, Imaging spectrometry : basic principles and prospective

applications. Remote sensing and digital image processing. 2001, Dordrecht ; Boston:

Kluwer Academic Publishers. xxiii, 403 p.

http://asterweb.jpl.nasa.gov/documents.asp

27

[25] Rowan, L.C. and J.C. Mars, Lithologic mapping in the Mountain Pass, California area

using advanced spaceborne thermal emission and reflection radiometer (ASTER) data.

Remote sensing of Environment, 2003. 84(3): p. 350-366.

[26] Clark, R.N., et al., High spectral resolution reflectance spectroscopy of minerals. Journal

of Geophysical Research: Solid Earth (1978–2012), 1990. 95(B8): p. 12653-12680.

[27] Clark, R.N., et al., USGS digital spectral library splib06a. 2007: US Geological Survey

Reston, VA.

[28] Mars, J.C. and L.C. Rowan, ASTER spectral analysis and lithologic mapping of the

Khanneshin carbonatite volcano, Afghanistan. Geosphere, 2011. 7(1): p. 276-289.

[29] Rowan, L.C., M.J. Kingston, and J.K. Crowley, Spectral reflectance of carbonatites and

related alkalic igneous rocks; selected samples from four North American localities.

Economic Geology, 1986. 81(4): p. 857-871.

[30] Bowers, T.L. and L.C. Rowan, Remote mineralogic and lithologic mapping of the ice river

alkaline complex, British Columbia, Canada, using AVIRIS data. Photogrammetric

Engineering and Remote Sensing, 1996. 62(12): p. 1379-1386.

[31] Bedini, E., Mapping lithology of the Sarfartoq carbonatite complex, southern West

Greenland, using HyMap imaging spectrometer data. Remote Sensing of Environment,

2009. 113(6): p. 1208-1219.

[32] Le Maitre, R.W., et al., Igneous rocks: a classification and glossary of terms:

recommendations of the International Union of Geological Sciences Subcommission on the

Systematics of Igneous Rocks. 2002: Cambridge University Press.

[33] Le Bas, M.J., Sovite and alvikite; two chemically distinct calciocarbonatites C1 and C2.

South African Journal of Geology, 1999. 102(2): p. 109-121.

[34] Gittins, J. and R.E. Harmer, What is ferrocarbonatite? A revised classification. Journal of

African Earth Sciences, 1997. 25(1): p. 159-168.

[35] Lee, W.-J., et al., Calciocarbonatite and magnesiocarbonatite rocks and magmas

represented in the system CaO-MgO-CO2-H2O at 0.2 GPa. Mineralogy and Petrology,

2000. 68(4): p. 225-256.

[36] Hay, R.L., Holocene carbonatite-nephelinite tephra deposits of Oldoinyo Lengai,

Tanzania. Journal of volcanology and geothermal research, 1989. 37(1): p. 77-91.

[37] WOOLLEY, A.R., et al., Extrusive Carbonatites from the Uyaynah Area, United Arab

Emirates. Journal of Petrology, 1991. 32(6): p. 1143-1167.

28

[38] Von Eckermann, H., The alkaline district of Alnö Island (Alnö alkalina område). 1948: A.

b. Kartografiska institutet.

[39] Hode Vuorinen, J., The Alnö alkaline and carbonatitic complex, east central Sweden-a

petrogenetic study. 2005.

[40] Dawson, J., The geology of oldoinyo lengai. Bulletin Volcanologique, 1962. 24(1): p. 349-

387.

[41] McInnes, B. and P. Wyllie, Scapolite formation and the production of nephelinitic melts

during the subduction of carbonated basalt. EOS Transactions, American Geophysical

Union, 1992. 73(43): p. 637.

[42] Nichols, G.T., P.J. Wyllie, and C.R. Stern, Subduction zone melting of pelagic sediments

constrained by melting experiments. Nature, 1994. 371(6500): p. 785-788.

[43] Lee, W.-J. and P.J. Wyllie, Petrogenesis of Carbonatite Magmas from Mantle to Crust,

Constrained by the System CaO–(MgO+ FeO*)–(Na2O+ K2O)–(SiO2+ Al2O3+ TiO2)-

CO2. Journal of Petrology, 1998. 39(3): p. 495-517.

[44] Le Bas, M., Nephelinites and carbonatites. Geological Society, London, Special

Publications, 1987. 30(1): p. 53-83.

[45] Jones, A.P., M. Genge, and L. Carmody, Carbonate melts and carbonatites. Rev Mineral

Geochem, 2013. 75(1): p. 289-322.

[46] Woolley, A.R. and B.A. Kjarsgaard, Carbonatite occurrences of the world: map and

database. 2008: Geological Survey of Canada.

[47] Bailey, D., Carbonate magmas. Journal of the Geological Society, 1993. 150(4): p. 637-

651.

[48] Bell, K. Carbonatites : genesis and evolution. London: Unwin Hyman.

[49] Bell, K., B. Kjarsgaard, and A. Simonetti, Carbonatites—into the twenty-first century.

Journal of Petrology, 1998. 39(11-12): p. 1839-1845.

[50] Gittins, J., The origin and evolution of carbonatite magmas. Carbonatites: genesis and

evolution. Unwin Hyman, London, 1989: p. 580-600.

[51] Xu, C., et al., Carbonatites in China: A review for genesis and mineralization. Geoscience

Frontiers, 2010. 1(1): p. 105-114.

29

[52] Lippard, S.J., et al., The geology of the Dibba zone, northern Oman mountains; a

preliminary study. Journal of the Geological Society, 1982. 139(1): p. 59-66.

[53] Nasir, S. and R. Klemd, New carbonatite occurrences along the Hatta transform fault zone

(northern Oman mountains, United Arab Emirates). Journal of African Earth Sciences,

1998. 27(1): p. 3-10.

[54] Bell, K., Radiogenic isotope constraints on relationships between carbonatites and

associated silicate rocks—a brief review. Journal of Petrology, 1998. 39(11-12): p. 1987-

1996.

[55] Verplanck, P.L. and B.S. Van Gosen, Carbonatite and alkaline intrusion-related rare earth

element deposits–A deposit model. 2011, US Geological Survey.

[56] Winter, J.D., Principles of igneous and metamorphic petrology. 2014: Pearson.

[57] Pell, J. and T. Höy, Carbonatites in a continental margin environment—the Canadian

Cordillera. Carbonate—genesis and evolution. Unwin Hyman, London, 1989: p. 200-220.

[58] Glennie, K.W., K.W. Glennie, and K.W. Glennie, The geology of the Oman Mountains: an

outline of their origin. 2005: Scientific Press Bucks.

[59] Robertson, A., et al., Evolution of the Arabian continental margin in the Dibba zone,

northern Oman Mountains. Geological Society, London, Special Publications, 1990. 49(1):

p. 251-284.

[60] Long, K.R., et al., The principal rare earth elements deposits of the United States: a

summary of domestic deposits and a global perspective. 2012: Springer.

[61] Oxides, R.E., Rare Earth Elements—Critical Resources for High Technology.

[62] Chen, Z., Global rare earth resources and scenarios of future rare earth industry. Journal

of Rare Earths, 2011. 29(1): p. 1-6.

[63] Cullers, R.L. and J.L. Graf, Rare earth elements in igneous rocks of the continental crust:

predominantly basic and ultrabasic rocks, in Rare earth element geochemistry. 1984,

Elsevier Amsterdam. p. 237-274.

30

2. CHAPTER 2

CARBONATITE IDENTIFICATION USING LANDSAT 8 VNIR-SWIR BANDS AT AL-

UYAYNAH, UAE

Abstract

Carbonatites, which are significant hosts of economically important rare-earth elements

(REE), are challenging to locate as they may crop out as small bodies with indistinct field

characteristics. Here we examine the spectral reflectance characteristics of a previously well-

exposed carbonatite that is located in the northeastern part of the United Arab Emirates (UAE).

We focus on the VNIR-SWIR bands between 0.45 to 2.2 µm, given this range is sensitive to the

influence of carbonatite minerals in the reflectance spectra. The Spectral Angle Mapper (SAM)

algorithm was utilized to identify carbonatites using the endmembers that were extracted from a

Landsat-8 OLI image. The SAM digital map displays a “good” result, with a calculated kappa

coefficient of 0.58, and an overall accuracy of about 67%. Our results therefore suggest that it is

possible to map and assess carbonatite distribution from medium spatial resolution (Landsat 8

OLI) images for exploration purposes, but the technique suffers from accuracy issues that may

negatively impact its usage for quantitative assessment of resource potential.

31

1. Introduction

Carbonatites are one of the most significant hosts for economically important Rare Earth

Elements (REE) such as neodymium (Nd), cerium (Ce), europium (Eu), and lanthanum (La) [6].

Identification of these carbonatites is often a challenge, as they commonly appear as smaller bodies

within larger-scale features [6, 7]. Remote sensing provides an opportunity to identify carbonatite

deposits, but the techniques applied require appropriate spectral resolution and the capability to

identify the spectrally significant features associated with carbonatites. Here we examine the

feasibility of using Landsat 8 OLI images for carbonatite detection by examining a well-exposed

carbonatite occurrence in the United Arab Emirates (UAE).

Remote sensing techniques have been a standard tool in mineral exploration over the past

four decades. Initially, these techniques were applied as a rapid and cost effective way to study the

chemical variability of surface rock types in order to map the lithological variation [8, 9]. Modern

remote sensing sensors have an improved spectral resolution that is useful in distinguishing

between different rock groups and mineral types [10, 11]. Specifically, it has been demonstrated

that a broad range of minerals, such as clays, carbonates, and hydroxides can be indicated by the

use of remote sensing spectral analysis (e.g. [12]).

Multispectral satellite sensors can cover the VNIR (visible/ near infrared), SWIR (short

wave infrared), and TIR (thermal infrared) bands, which are important in mineral exploration [13].

The VNIR and SWIR bands are typically a valuable source of information about transition metal

absorption, specifically iron minerals, and also are sensitive to some of the rare-earth elements

32

(REE) [11, 12]. In particular, the wavelength region (0.45 to 2.5 µm) of the reflectance spectrum

is specifically used to indicate the composition of minerals based on absorption features [12, 14].

Carbonatites, which contain more than 50% carbonate minerals, exhibit a particularly

strong CO32- absorption feature between 1.65 and 2.2-2.33 μm [15]. For carbonatites enriched in

iron (ferro-carbonatites) there is an additional absorption feature between 1.0 and 1.3 μm [16].

While these absorption features are distinctive, they are not uniquely diagnostic of carbonatites;

other carbonate rocks such as limestone and marble also exhibit similar absorbance peaks. In

addition, REEs, which can be a significant component in carbonatite magmas, exhibit absorption

in the wavelength range between 0.5 and 0.9 μm that is dominantly attributed to electronic

transitions in Nd3+ [17, 18] By combining these absorption features we may resolve lithological

ambiguities and allow for distinction between igneous (i.e. carbonatites) and sedimentary (e.g.

limestone) carbonate rocks.

The absorption spectral features of carbonatites have been used in previous research to

create maps of carbonatite complexes that are detectable from aircraft sensors. AVIRIS (Airborne

Visible / Infrared Imaging Spectrometer) images were used to distinguish the Ice River Carbonatite

complex, and showed minor absorption features located at 0.58, 0.74, 0.80, and 0.87 μm, which is

characteristic of Nd3+ [19]. Recent studies of the Sarfartoq carbonatite in Greenland by Bedini and

Tukiainene were conducted using spectral mixing methods to distinguish between carbonatite,

fenite, and hematised surface rock, with particular attention to the Nd3+ reflectance characteristics

[20].

33

In this study, we use Landsat 8 spectral data [21] obtained within the Visible/Near Infra-

Red (VNIR) and Short Wave Infra-Red (SWIR) bands to map extrusive carbonatite occurrences

at Al-Uyaynah in the northwestern Oman Mountains. We evaluate the utility of Landsat 8 OLI

data and the SAM classification method for mapping carbonatite distribution. We find that this

method has utility for exploration purposes but suffers from accuracy issues that may negatively

impact its usage for quantitative assessment of resource potential.

2. Al-Uyaynah

The study area for this project is the Al-Uyaynah carbonatite field of the United Arab

Emirates, which is located approximately 18 km to the southwest of Dibba city (Figure. 2.1). The

Al-Uyaynah carbonatite is mapped as part of metamorphic sole of Samail ophiolite, which contains

exposures of sub-aerially erupted carbonatites within oceanic metasediments [7]. The Al-Uyaynah

carbonatite has previously been interpreted as an accreted volcanic island predating the Samail

ophiolite [7]. The exposures (about 10-50 m) are an ideal location to undertake a remote sensing

study (Figure 2.1).

The Al-Uyaynah carbonatite is composed of two typical magma types (calcio- and ferro-

carbonatite) (Figure 2.2), which are also broadly representative of most global carbonatite

occurrences. The carbonatite complex is comprised of ~10-50 m thick layers with lapilli, which

require a sub-aerial origin [7]. The main mineral phase in the calcio-carbonatites is calcite.

However, these carbonatites also contain abundant apatite, and opaque minerals such as magnetite.

The typical composition of these lapilli is 40% calcite, 20-30% apatite and 20-40% opaque

34

minerals [7]. The existing whole rock geochemical data show that ferro-carbonatites have higher

modal abundances of silicates in comparison calcio-carbonatites. Representative minerals of ferro-

carbonatites are allanite (Ce), amphibole (alkali) and occasionally strontian barite. Allanite in

ferro-carbonatites usually occurs as crystal accumulations. These accumulations concentrate REEs

and result in trace elements heterogeneously distributed within the rock [7].

3. Methods

3.1. Landsat 8 data calibration

The Landsat 8 OLI multispectral images were obtained from the USGS as a Level1 T-

Terrain Corrected (orthorectified) image [21]. This image was acquired on May 27th, 2014 with

2.87 % cloud cover, which is centered away from the area of interest for this study. Overall image

quality is 9 out of 9 on a NASA scale. The Landsat OLI sensor acquires images in the spectral

range of 0.43 - 2.29 µm (VNIR-SWIR) with a 30-meter pixel resolution, and within the

panchromatic band at a 15-meter resolution. The standard data product of Landsat 8 is

georeferenced in Universal Transverse Mercator (UTM) map projection using a World Geodetic

System datum (WGS-84) [21]. The pre-processing steps applied to the Landsat 8 VNIR-SWIR 30

meter resolution datasets were: (1) conversion of the raw digital image to sensor radiance using

Landsat Calibration data and (2) application of the QUick Atmospheric Correction (QUAC)

model, that is accessible in the Environment for Visualizing Images (ENVI) software [22], in order

to eliminate the atmospheric contributions.

35

Figure 2.1. A map of the northern eastern of UAE indicating the location of the Al-Uyaynah area and general

geological units, modified from [1, 2], and Landsat 8 natural colored image of Al-Uyaynah area shows the

location of study area (right).

36

Figure 2.2. A field view of Al-Uyaynah carbonatites (fig. 1), (a) shows the ferro-carbonatites outcrops of the

northern end of the study area, (b, and c) show a close view of ferro-carbonatites. (d) Shows the extruded calcio-

carbonatite, and (e, f) illustrates a close view of calcio-carbonatite.

37

3.2. Landsat data analysis

The Minimum Noise Fraction (MNF) method is a linear transformation function utilized

to differentiate noise from data and to diagnose dimensionally coherent spectral data [22, 23]. A

forward MNF transform was applied to the image by inspecting eigenvalues and the seven MNF

gray bands (Figures 2.3, 2.4.1), in order to identify which bands had coherent data. We utilized the

Pixel Purity Index (PPI) to find the spectrally extreme pixels, or pure, pixels within the image [22,

23]. The PPI is calculated by frequently projecting a n-D (n-dimensional) scattergraph on an

indiscriminate unit vector [22, 23]. The PPI functions were applied on an MNF image with 20,000

iterations, and then the results were visualized using n-D visualizer (n-D scatter plots). We selected

pixels of the unit vector to be the spectral reflectance end-members.

3.3. Field work and ASD spectral measurements

Rock samples were collected during a field expedition in December 2010. These samples

were analyzed using an Analytical Spectral Device (ASD) spectroradiometer at the University of

Toledo, Ohio, U.S.A. Spectral measurements were performed in a laboratory using an ASD

FieldSpec-3 Hi-Res Spectroradiometer. Illumination was provided by a 1000W 3200K Quartz

Halogen lamp. A 10 degree foreoptic was employed. Spectra were acquired for each sample. All

samples were normalized to a Spectrlon lab standard with well-characterized spectral properties.

Lab spectra were resampled to OLI bands for comparison using the spectral resampling function

in ENVI [22].

38

3.4. The spectral angle mapper (SAM)

We selected the Spectral Angle Mapper (SAM) algorithm as our mechanism for image

classification as it has previously been successfully applied to problems involving mineral

identification and geological mapping [e.g.[4, 24]]. SAM is a physical method used to classify an

image by comparing and matching the angle between the pixel vector and an end-member spectral

vector in n-D space [22, 23, 25]. A smaller angle shows a better match to spectral endmembers;

pixels exceeding the maximum angle of the end member’s threshold are not counted in the

classification [26]. The technique is effective given that it does not require any assumptions for

numerical/statistical distribution of input- information in carrying out a classification [27].

Furthermore, the technique suppresses the impact of effects caused by topographic shading [25].

In this study, our image was classified based on the SAM algorithm using the extracted

endmembers’ reflectance spectra. In the output each image represents one end member (see more

details in results section).

4. Results

4.1. The image processing results

Landsat 8 OLI bands, which cover the VNIR and SWIR spectral range (0.433–2.95 µm),

were chosen to perform the MNF transformation. The results of this transformation included

eigenvalues and eigenimages (MNF images), which were used to assess the dimensionality of the

data by comparing MNF images with MNF eigenvalues (Figure 2.3). The larger eigenvalues

39

Figure 2.3. MNF eigenvalue plot for study area (Al-Uyaynah). All MNF bands have eigenvalues values greater

than 1. The lowest band (7) has eigenvalue equal to 1.23, which means it contains coherent data. However, the

first three bands contain more spatially coherent information than the others.

40

Figure 2.4. MNF images of Landsat 8 VNIR-SWIR bands for study area. Images 1-7, display a single gray

band, while image 8 shows a combination of first three MNF bands (1, 2 and 3 as RGB). Image 9 shows

natural color image (Landsat Bands 3, 2, 1 (RGB)) for comparison (Not MNF image).

41

Figure 2.5 Show a 3-D Scatter plot of the end-members data cloud (image bands 1-3 were used as plot

axes). The identical rock units were characterized by the same color.

42

(greater than 1) contain the greater portion of inherent information, and bands with eigenvalues

close to 1 contain noise [22]. All MNF bands used have eigenvalues values greater than 1.23;

however, the first three bands (1-3) contain more spatially coherent information than the other

four bands (4-7) as shown in Figures. 2.3 & 2.4.

4.2. Carbonatite endmember spectral characteristics

4.2.1. The ferro-carbonatite 

The ferro-carbonatite image endmember spectra extracted from the OLI image showed

absorption features of calcite (CO32- ion) at 1.6 µm as a weak absorption, which then descended

gradually to reach the Landsat OLI sensor limit at 2.2 µm (Figure 2.8). The laboratory (ASD)

spectra of ferro-carbonatite showed that calcite has a stronger absorption in the window between

2.2 to 2.33 µm (Figure 2.5, and 2.7b). Ferro-carbonatites have a higher modal abundance of

silicates, which manifest as dark inosilicate minerals (alkali amphibole) [7]. The hydroxyl ion in

Al-OH and Mg-OH of amphibole minerals displays absorption characteristics at 1.4 µm and 2.2-

2.3 µm [13]. The presence of alkali amphibole minerals in the ferro-carbonatite

image endmember spectra results in an increase in values of the absorption near 2.1 µm and

reaches the strongest absorption at 2.2 µm; this is more obvious in the ASD spectra of the ferro-

carbonatite (Figure 2.7b) than in the spectra extracted from the Landsat image (Figure 2.8). The

ASD spectra of ferro-carbonatite showed that alkali amphibole minerals have 1.4, 2.15, 2.16, and

2-2.2 µm absorptions. The high concentration of ferric iron (Fe3+) and REE in minerals within the

43

Figure 2.6. ASD Resampled spectra to Landsat 8 of (a) ferro-carbonatite and (b) calcio-carbonatite.

Figure 2.7. A laboratory reflectance spectra (ASD) of samples, (a) ferro-carbonatite and (b) calcio-carbonatite (left). (a) A zoom of spectral region

0.5-0.6 µm of laboratory reflectance spectra (ASD) of REE absorption are near 0.589, 0.567, and 0.56 µm, and (b) A zoom of spectral region 2000-

2400 nm of laboratory reflectance spectra (ASD) calcite (CO3, MgO) REE absorption are near 2.32-2.35µm and calcite has weak absorption at 1.6

µm.

a b

44

rocks lowers the reflectivity or increases the absorption over the visible/very near infrared (VNIR)

[28]. The absorption in the VNIR bands of the ferro-carbonatite image spectrum from 0.865-

0.9 µm could be because of the existence of ferrous-ions in the ferro-carbonatite, and ferric-ions

(Fe3+) in its weathered surface. Apatite (Nd3+) absorption behavior is present as a weak absorption

near 0.589, 0.567, and 0.560 µm in the ferro-carbonatite ASD spectra (Figure 2.7a). This

absorption is not clear in the image endmember spectrum but we have noted a lower reflectivity at

0.589, and 0.567 µm. Ferro-carbonatites are rich in allanite (Ce), resulting in low absorption at

0.45 - 0.5 µm in both the ASD and image endmember spectra.

4.2.2. The calcio-carbonatite 

The calcio-carbonatite image endmember has absorption features of calcite that fall

gradually from 2.1-2.2 µm (sensor limit). Calcio-carbonatites contain fewer silicate minerals but

more calcite than ferro-carbonatites, reflecting stronger absorption characteristics between 2.0 and

2.2 µm. The ASD spectra of calcio-carbonatites show a strong absorption between 2.2-2.33 µm

(Figures 2.5, and 2.7b). Alkali amphibole is present as a secondary mineral in calcio-carbonatite

and has a weak vibrational absorption near 1.4 and 2.2 µm in the ASD spectra, but is not detected

at 1.4 µm in the calcio-carbonatite image endmember due to the atmospheric H2O vapor absorption

bands (0.9-1.5 µm).

45

…………………………………………………………

Figure 2.8. Landsat 8 image endmembers spectra of (a) calcio-carbonatite in red color and (b) ferro-

carbonatite in green color; within VNIR-SWIR bands.

46

4.2.3. The difference between ferro/calcio-carbonatite endmembers

Ferro-carbonatite and calcio-carbonatite have similar behaviors or absorption within the

visible band, between 0.865-0.9 µm, due to the presence of iron oxides on the surfaces of

weathered outcrops. Furthermore, the presence of calcite in both calcio- and ferro-carbonatites

results in an increase in absorption at 2.1 and 2.2 µm. Some distinction between the two rock types

is evident due to the greater percentage of calcite within the calcio-carbonatites, which manifests

as more absorption at 2.2 µm. REE absorption features make ferro-carbonatite distinguishable

from calcio-carbonatite in ASD spectra. The weak absorptions near 0.605, 0.589, 0.567, and 0.45

-0.5 µm of the ferro-carbonatite ASD is due to the high abundance of REE-bearing minerals

(allanite (Ce), and apatite).

4.2.4. Other endmembers

Within the study area there are a number of other (non-carbonatite lithologies (Fig. 9)

which must be assessed in order to fully classify our Landsat image:

(i) The alkaline volcanic unit. This unit is dominated by deformed basaltic alkali volcanic

rocks (sub/green schist facies). Alkali basalts typically have high

concentrations of Na2O [28], which causes a weak absorption close to 0.65 µm [28]. The

endmember of alkaline rocks shows this same weak absorption close to 0.65-0.68 µm.

Furthermore, the alkali basalt endmember exhibits aluminum hydroxide (Al-OH) absorption

characteristics from 2.1 to 2.2 µm.

47

(ii) The metavolcanic unit. This unit’s n-D spectrum displays absorption features of

phlogopite, in which it shows absorption from 1.6 to 2.2 µm, and within the visible band (0.45

µm) (Figure 2. 9). These spectral features are due to absorption by the transition pair of Fe2+ and

Fe3+ ions in phlogopite [28]. The OH lattice in magnesium hydroxide (Mg-OH) and aluminum

hydroxide (Al-OH) of metavolcanic ‘phlogopite’ causes electronic and vibrational transition

features responsible for the reflected energy consumption and absorption in the metavolcanic unit

n-D spectra at 1.6, which decreases gradually to 2.2 µm.

(iii) The ophiolite complex. The average image spectra from the ophiolite complex has

several absorption bands between 1.6-2.2µm. It is, however, absorbed strongly in the area of 2.2

µm due to the high abundances of silicates (e.g., olivine).

(iv) The red chert unit. This unit is rich in silicate minerals (quartz) and contains iron

oxides. The red chert endmember spectrum has silicate absorption near 2.2 µm, and iron oxide

absorption characteristics from 0.865 µm to 0.443 µm (Figure 2.9).

(v) Schist formation. Chlorite (Mg-Fe-O) and muscovite (Al-OH) are the major minerals

in these rocks, which has created an absorption feature in the VNIR band (0.4-0.09 µm). In

addition, a weak Al-OH absorption at 2.1-2.2 µm, caused by muscovite, is evident in the schist

image endmember.

48

Figure 2.9. The image n-D spectra were extracted from Landsat OLI image used in SAM classification.

49

4.3. The SAM classification results

The Spectral Angle Mapper algorithm (SAM) algorithm was utilized in the calculation of

the spectral likeness between endmembers and image spectra [23, 25, 26]. SAM was applied to

the Landsat 8 image, using ENVI 5.1 software, in order to indicate and map both types of

carbonatites and other rock units around it based on extracted reflectance endmembers. Afterward,

the results of SAM method were compared to the field ground truth data and the geological map

in order to generate the accuracy assessment, which was calculated using a confusion-matrix

method [22, 29]. The performance of the SAM filtering method displays overall a “good”

differentiation between ferro/calcio-carbonatite and other units (Figure 2.10, 2.11, and 2.12). The

accuracy assessment of the SAM shows overall accuracy with 66.92%. Kappa Coefficient is 0.58,

which falls into the ‘good’ category [3, 4] (Table 1). We must emphasize that this assessment is

for the image as a whole– accuracy in terms of carbonatite differentiation was significantly more

robust as discussed below.

5. Discussion

Comparing the distribution of spectrally defined lithological units to geologically mapped

terrains is central to the assessment of any exploration remote sensing technique. Our area contains

seven spectrally-defined endmembers which outcrop in a complex pattern as defined in the existing

geological map. Our overall accuracy assessment of 66.92% is a composite of the mismatch

between spectrally predicted and observed lithologies at any point for all these units. However,

50

Lower Bound Degree of Agreement Upper Bound Lower Bound

<0.05 No 0.05 <0.05

0.05 Very Poor 0.20 0.05

0.20 Poor 0.40 0.20

0.40 Fair 0.55 0.40

0.55 Good 0.70 0.55

0.70 Very Good 0.85 0.70

0.85 Excellent 0.99 0.85

0.99 Perfect 1.00 0.99

Table 2.1. The values of threshold that used for dividing the results of statistical kappa coefficient degrees

of agreement. From [3-5]. The values close or equal to 1.0 specify perfect agreement; while any values

between 0.85-0.99 show excellent agreement, and any kappa coefficient values of between 0.75-0.85 would

be a very good agreement. The lesser degree of agreement were classified as: good if the kappa coefficient

values were between 0.4-0.55; fair agreement between 0.40-0.55 values; and kappa coefficient values

between 0.4-0.05 or fewer show poor to no agreement. The SAM map accuracy assessment displays overall

accuracy with 66.92%. Kappa Coefficient is 0.58, which falls into the ‘good’ category.

51

Figure 2.10. SAM n-D classification map of Al-Uyaynah area shows the extrusive carbonatites and the

lithology unites.

52

11111

Figure 2.11. Geological map of Al-Uyaynah area shows the extrusive carbonatites (modified from British

geological survey, 2006).

53

accuracy values vary significantly from unit to unit. To fully assess the class distribution accuracy

and the potential reasons for error, below we discuss each mapped class.

5.1. Distribution of calcio/ferro-carbonatites units

Utilizing the SAM method, we were able to distinguish both calcio and ferro-carbonatites

from other major rock units. The discrimination of calcio-carbonatites (82.50 % of class accuracy)

was better than ferro-carbonatites with 70.72 % accuracy (Table 2), and Figure 2.12. The spatial

distribution of calcio/ferro-carbonatites closely matched the reference geological map. Despite

these good correlations, the 17.5 % mismatch for calcio-carbonatites and 29.28% for ferro-

carbonatites requires examination. We found that this inaccuracy resulted from: (1) The non-

detection of some calcio-carbonatites due to the variable size range of carbonatite bodies

(approximately 10-50 meters). (2) Sub pixel contamination, caused by the presence of multiple

rock types within 30 square meter pixel size of the Landsat 8 OLI image, may confound the SAM

algorithm. (3) Spectral similarity between carbonatite units, the alkaline volcanic formation, and

metavolcanic rocks in close geographic proximity can further complicate pixel discrimination.

Carbonatites and alkaline volcanic rocks both contain alkaline amphibole, while surrounding

metamorphic rocks contain micas (chlorite and phlogopite) that may also be found in carbonatites.

These compositional and spectral similarities between carbonatites and surrounding rock types led

the SAM algorithm to map these quite distinct lithological units as one contiguous spectrally-

defined unit (Figure. 2.10, 12).

54

5.2. Distribution of other non-carbonatite units

The Ophiolite complex, as illustrated using the SAM method, matched the available

geological maps with approximately 94.5% (Figure. 10, 11, and Table 2). The SAM map of

metavolcanic unit is about 50% accurate in comparison to the metavolcanic formation in the

western part of the geologic map. Due the spectral likeness, the SAM algorithm mapped some of

metavolcanic units as alkaline volcanic rocks (Figure 2.10, and 2.12).

The alkaline volcanic unit, which is dominated by deformed basaltic alkali volcanic rocks

(sub/green schist facies), was mapped using the SAM algorithm. A comparison of the geological

maps with the alkaline distribution shows about 20% of it were accurately classified.

The SAM map shows red chert in the areas between the ophilite and alkaline volcanic units,

which matches the geological map with 32%. However, in the south portion of our study area, the

red chert formation was classified by the SAM algorithm as Metavolcanic (Figure 2.10, and 2.12).

This was due to the high proportion of silicate minerals in both units in this region.

The schist formation, was clearly mapped using the SAM method, and shows a strong

correlation (94.5%) between the spectrally identified schist endmember and the mapped schist unit

in the north-western part of the geologic map.

55

6. Conclusion

This paper offers a review of Landsat 8 OLI data and SAM algorithm validation for use in

mapping of carbonatite outcrops. Landsat imagery remains among the most popular remote

sensing datasets for geological mapping and mineral exploration. Our results, which provide an

assessment of the new Landsat 8 capabilities in a well-studied carbonatite setting, are therefore of

particularly broad utility to the geologic remote sensing community.

The primary rocks of the study area (extrusive carbonatites) were well-represented by the

calculated endmembers. Ferro-carbonatite rocks were mapped with an accuracy of 70.72%, with

better results from the calcio-carbonatite endmember, which has an accuracy of 82.5%. Challenges

in the discrimination of ferro-carbonatite with the SAM algorithm arose because these highly

deformed bodies are relatively narrow (generally only a few meters in width) and are located close

to the alkali mélange unit. Mixtures of the units can prove problematic for the SAM algorithm,

which does not take the sub-pixel mixing into account, and Landsat 8, which has medium spatial

resolution and limited spectral range in the SWIR bands (1.57 - 1.65 and 2.11 - 2.29 µm). These

limitations produce ambiguity in the observations and make it difficult to clearly distinguish

carbonatite from other rocks. Further complicating factors were the iron oxide minerals that were

found in weathered outcrops – these coatings resulted in some similarities between calcio-

carbonatite and ferro-carbonatite weathered surface spectra.

In conclusion, our comparison of the medium resolution geological map of the Al-Uyaynah

extrusive carbonatite with the results of the SAM algorithm showed “good” results. The

56

inaccuracies noted between mapped and spectrally-inferred lithological units, result from the

spatial resolution limitations of Landsat 8, and the fact that endmember selection was complicated

by sub-pixel mixing between variable rock cover types. Despite these issues, we have

demonstrated that this method proves effective for geologic mapping and mineral general

exploration purposes.

57

APPENDIX

58

Figure 2.12. A Comparison between the rock units’ distribution that classified correctly, in percent, of SAM map based on the results of the confusion matrix

accuracy assessment. SAM method gave, overall, a good result after comparing SAM map to the field ground truth regions.

59

REFERENCES

60

REFERENCES

[1] W. B. Ryan, S. M. Carbotte, J. O. Coplan, S. O'Hara, A. Melkonian, R. Arko, et al.,

"Global Multi‐Resolution Topography synthesis," Geochemistry, Geophysics,

Geosystems, vol. 10, 2009.

[2] M. Tarapoanca, P. Andriessen, K. Broto, L. Chérel, N. Ellouz-Zimmermann, J.-L.

Faure, et al., "Forward kinematic modelling of a regional transect in the Northern

Emirates using geological and apatite fission track age constraints on paleo-burial

history," Arabian Journal of Geosciences, vol. 3, pp. 395-411, 2010.

[3] R. A. Monserud and R. Leemans, "Comparing global vegetation maps with the Kappa

statistic," Ecological modelling, vol. 62, pp. 275-293, 1992.

[4] F. A. Kruse and S. L. Perry, "Mineral mapping using simulated Worldview-3 short-

wave-infrared imagery," Remote Sensing, vol. 5, pp. 2688-2703, 2013.

[5] D. Cooper, M. Ali, and M. Searle, "Structure of the northern Oman Mountains from

the Semail Ophiolite to the Foreland Basin," Geological Society, London, Special

Publications, vol. 392, pp. 129-153, 2014.

[6] D. Singer, "Descriptive model of carbonatite deposits," Mineral deposit models: US

Geological Survey Bulletin, vol. 1693, p. 51, 1986.

[7] A. R. Woolley, M. W. C. Barr, V. K. Din, G. C. Jones, F. Wall, and C. T. Williams,

"Extrusive Carbonatites from the Uyaynah Area, United Arab Emirates," J.

Petrology, vol. 32, pp. 1143-1167, December 1, 1991 1991.

[8] G. R. Hunt, J. W. Salisbury, and C. J. Lenhoff, "Visible and near-infrared spectra of

minerals and rocks; IV, Sulphides and sulphates," Modern Geology, vol. 3, pp. 1-14,

1971.

[9] F. v. d. Meer and S. M. d. Jong, Imaging spectrometry : basic principles and

prospective applications. Dordrecht ; Boston: Kluwer Academic Publishers, 2001.

[10] Application of remote sensing to hydrology; final technical report. Huntsville, Ala.:

IBM Federal Systems Division, reproduced by the National Technical Information

Service, 1973.

[11] L. C. Rowan and J. C. Mars, "Lithologic mapping in the Mountain Pass, California

area using advanced spaceborne thermal emission and reflection radiometer (ASTER)

data," Remote sensing of Environment, vol. 84, pp. 350-366, 2003.

61

[12] R. N. Clark, T. V. King, M. Klejwa, G. A. Swayze, and N. Vergo, "High spectral

resolution reflectance spectroscopy of minerals," Journal of Geophysical Research:

Solid Earth (1978–2012), vol. 95, pp. 12653-12680, 1990.

[13] C. Elachi and J. J. Van Zyl, Introduction to the physics and techniques of remote

sensing vol. 28: John Wiley & Sons, 2006.

[14] G. R. Hunt and J. W. Salisbury, "Visible and near infrared spectra of minerals and

rocks. II. Carbonates," Modern Geology, vol. 2, pp. 23-30, 1971.

[15] J. C. Mars and L. C. Rowan, "ASTER spectral analysis and lithologic mapping of the

Khanneshin carbonatite volcano, Afghanistan," Geosphere, vol. 7, pp. 276-289, 2011.

[16] F. Van der Meer, "Spectral reflectance of carbonate mineral mixtures and

bidirectional reflectance theory: Quantitative analysis techniques for application in

remote sensing," Remote Sensing Reviews, vol. 13, pp. 67-94, 1995/08/01 1995.

[17] R. N. Clark, Spectroscopy of rocks and minerals and principles of spectroscopy.

United States (USA): John Wiley & Sons, New York, NY, United States (USA),

1999.

[18] L. C. Rowan, M. J. Kingston, and J. K. Crowley, "Spectral reflectance of carbonatites

and related alkalic igneous rocks; selected samples from four North American

localities," Economic Geology, vol. 81, pp. 857-871, 1986.

[19] T. L. Bowers and L. C. Rowan, "Remote mineralogic and lithologic mapping of the

ice river alkaline complex, British Columbia, Canada, using AVIRIS data,"

Photogrammetric Engineering and Remote Sensing, vol. 62, pp. 1379-1386, 1996.

[20] E. Bedini, "Mapping lithology of the Sarfartoq carbonatite complex, southern West

Greenland, using HyMap imaging spectrometer data," Remote Sensing of

Environment, vol. 113, pp. 1208-1219, 2009.

[21] NASA Landsat Program, "Landsat 8 scene LC81600422013088LGN01," SLC-Off,

USGS, Sioux Falls, May 06, 2014, 3:46:30 PM 2014.

[22] E. V. I. Solutions. (2014). Exelis ENVI online manual. Available:

http://www.exelisvis.com/docs/home.html.

http://www.exelisvis.com/docs/home.html

62

[23] F. A. Kruse, J. W. Boardman, and J. F. Huntington, "Comparison of airborne

hyperspectral data and EO-1 Hyperion for mineral mapping," Geoscience and Remote

Sensing, IEEE Transactions on, vol. 41, pp. 1388-1400, 2003.

[24] G. Girouard, A. Bannari, A. El Harti, and A. Desrochers, "Validated spectral angle

mapper algorithm for geological mapping: comparative study between QuickBird and

Landsat-TM," in XXth ISPRS Congress, Geo-Imagery Bridging Continents, Istanbul,

Turkey, 2004, pp. 12-23.

[25] F. Kruse, A. Lefkoff, J. Boardman, K. Heidebrecht, A. Shapiro, P. Barloon, et al.,

"The spectral image processing system (SIPS)—interactive visualization and analysis

of imaging spectrometer data," Remote sensing of environment, vol. 44, pp. 145-163,

1993.

[26] E. V. I. Solutions. (2014). Exelis ENVI online manual

[27] G. P. Petropoulos, K. P. Vadrevu, G. Xanthopoulos, G. Karantounias, and M.

Scholze, "A comparison of spectral angle mapper and artificial neural network

classifiers combined with Landsat TM imagery analysis for obtaining burnt area

mapping," Sensors, vol. 10, pp. 1967-1985, 2010.

[28] R. N. Clark, G. A. Swayze, R. Wise, K. E. Livo, T. M. Hoefen, R. F. Kokaly, et al.,

USGS digital spectral library splib06a: US Geological Survey Reston, VA, 2007.

[29] J. R. Jensen, Introductory digital image processing: a remote sensing perspective.

Upper Saddle River, N.J: Prentice Hall, 2005.

63

3. CHAPTER 3

A COMPARISON OF MAPPING METHODS AND SPECTRAL ANALYSIS TO

IMPROVE CARBONATITE MAPPING USING VNIR AND SWIR BANDS OF

ASTER DATA AT AL-UYAYNAH AREA, UAE

Abstract

This study demonstrates an evaluation of ASTER multispectral data analysis and

interpretation in the range of visible/near-infrared (VNIR) and Short Wave Infrared (SWIR)

and mapping methods for lithological mapping of carbonatites. Carbonatites, which are

usually rich in economically significant minerals/elements such as rare-earth elements

(REE), are part of the metamorphic sole of Samail ophiolite that is located in the northeastern

terrane of United Arab Emirates (Al-Uyaynah area). In this study, we have examined four

different algorithms to identify carbonatites: Binary Encoding Classification (BEC), Support

Vector Machine (SVM), Spectral Information Divergence (SID), and Spectral Angle Mapper

(SAM). These algorithms were applied to identify the integral correlation between

carbonatite and neighboring rock units. The results are shown as digital-based maps that

highlight pixels that correspond to particular endmember-spectra among different kinds of

class. Out of the methods we tested, the SVM distribution map displays the most robust

results; the agreement with field ground truth ROIs data is excellent with overall accuracy at

64

97.58%, and with a calculated kappa coefficient of 0.97. The SAM, SID, and BEC yield

overall accuracy between 81%-43% respectively and kappa coefficient between 0.75 for

SAM to 0.32 for BEC. To confirm our finding, the SVM mapping method was tested on

another carbonatite, the Hatta carbonatite, located in a region 60km south our main study

area (about 60 Km). Hatta shares the same petrogenetic history and major geochemical

characteristics as the Al-Uyaynah area. The result shows that SVM mapped the Hatta

carbonatites with an excellent overall result of 99% accuracy and a kappa coefficient of 0.99.

Our results, which over both carbonatite localities exhibit excellent results, highlight the

potential utility of the SVM technique applied to ASTER data for further carbonatite

exploration and mapping.

1. Introduction

This research is focused on examination of ASTER (Advanced Spaceborne Thermal

Emission and Reflection Radiometer) image information (Table. 1, and Figure 3.1) to map

the Al-Uyaynah carbonatite, a small area within the metamorphic sole of Samail ophiolite,

located within northeastern Oman’s mountains. Carbonatites are typically important hosts for

economically useful minerals/elements; such as rare earth elements (REE), which include

cerium (Ce), lanthanum (La), neodymium (Nd), and europium (Eu) [4]. Carbonatite detection

is often a challenge, as it commonly appears as narrow bodies within larger-scale features

[5]. To address this challenge, a more widespread use of spectral base sensors, with world-

wide coverage and proper wavelength resolution is required.

65

Figure 3.1. ASTER spectral transmittance of band pass filters were used in this research, modified from[13].

66

The application of remote sensing for mineral/element exploration and mapping has

expanded remarkably since the early 1960s. Initially, remote-sensing was used as an

affordable and quick method to analyze the chemical deviation of surface rocks types in order

to map the lithological variance within a region [6, 7]. Multi-spectral satellite sensors include

the visible/near-infrared (VNIR), shortwave infrared (SWIR), and thermal infra-red (TIR)

spectra that are crucial to mineral exploration [8, 9]. The Advanced Spaceborne Thermal

Emission and Reflection Radiometer (ASTER) acquires images in the (VNIR, SWIR, and

TIR) ranges-spectra, which are important in mineral exploration [10]. The ASTER sensor

can reveal more bands than the old sensors within the spectral range sensitive to minerals and

rocks at the earth’s surface [11]. Previous researchers pointed out that ASTER compares

favorably with the precision of other sensors and has an advantage for geological mapping

[12] by having better reflectance spectra, allowing for the identification of assemblages of

specific minerals or groups of minerals [7].

Contemporary remote-sensing detectors like ASTER have a greater spectral

bandwidth that is beneficial for the differentiation of rock bands and minerals [14, 15].

Mainly it has been shown that a wide array of minerals such as clays, carbonates, and

hydroxides could be recognized through the usage of remote-sensing spectral analysis

methods (e.g.[9]). The VNIR and SWIR bands’ data typically are an invaluable source of

information regarding transition elements’ absorption, particularly iron species like Fe2+ and

Fe3+ [8, 15, 16]. The area of the reflectance spectrum from 0.45 to 2.5 µm is important in

determining the composition of minerals on the basis of absorption characteristics [9, 15, 17].

67

Carbonatites, which are enriched with about 50% or more of calcite minerals, exhibit a

particularly strong absorption characteristic of the CO32- ion, which observed within the range

between 2.31-2.33 µm of the SWIR band [18].

The absorption features of iron-rich carbonatites are pronounced in the range of

wavelengths between 1.0-1.3 µm [19]. Iron and REEs, which may occur in carbonatites in

abundance, but are not typically found in most other carbonates, may be an important factor

in distinguishing carbonatites from other economically less valuable carbonate rocks. The

REE spectrum is largely related to electronic transitions in Nd3+, which is clearly shown by

absorption features in the wavelength ranges between 0.5 and 0.9 um [9, 20]. A mixture of

these absorption characteristics can serve as a foundation for the creation of carbonatite maps

through remote sensing. The absorption spectra attribute of carbonatites has already been

employed in previous investigations to generate maps of carbonatite complexes that are

detectable by aircraft sensors. AVIRIS (Airborne Visible / Infra-Red Imaging Spectrometer)

scenes were utilized to recognize the Ice River carbonatite complex, and revealed slight

absorption characteristics situated at 0.58, 0.74, 0.80, and 0.87 um, which suggests a

neodymium absorption feature [21]. Similarly, recent studies of the Sarfartoq carbonatite in

Greenland were conducted using a mix of spectral techniques to differentiate between

carbonatite, fenite, and hematised areas of exposed rock, with special focus on the Nd3+

reflectance feature [22]. Comparatively little work on carbonatite identification has been

attempted using satellite-based sensors.

68

The aims of this research are: (1) to investigate and assess the extent of the Al-

Uyaynah carbonatite by analyzing the spectral response utilizing ASTER visible near

infrared (VNIR) and short wave infrared (SWIR) bands, (2) to produce a more accurate

classification of the ASTER dataset of carbonatite distribution by utilizing both spectral and

spatial information, and (3) to assess ASTER system performance for carbonatite mapping.

2. Geologic background of study

2.1 Carbonatite location and outcrops

The area of interest for this research project is the Al-Uyaynah carbonatite of the

United Arab Emirates. This region is approximately 18 kilometers to the southwest of Dibba

city (Figure. 3.2). The Al Uyaynah carbonatite is mapped as a member of the metamorphic

sole of Samail ophiolite, which is comprised of exposures of sub-aerially erupted carbonatites

within oceanic metasediments [5]. The Al Uyaynah carbonatite has formerly been construed

as a volcanic accretionary island pre-dating the Samail ophiolite [5]. The outcrops are well-

exposed and are roughly 10-15 meters thick of many of adjacent layers in plan view, making

this region a useful locale to study remote sensing techniques applied to carbonatites (Fig.

2).

69

Figure 3.2. (a) Map of the northern eastern of UAE indicating the location of the Al-Uyaynah area and

general geological units, modified from [23, 24]. (b) ASTER 3, 2, and 1 natural color image.

70

Figure 3.3. A field view of Al-Uyaynah carbonatites (fig. 1), (a) shows the ferro-carbonatites outcrops of

the northern end of the study area, (b, and c) show a close view of ferro-carbonatites. (d) Shows the extruded

calcio-carbonatite, and (e, f) illustrates a close view of calcio-carbonatite.

71

2.2 Geochemistry of Al-Uyaynah Carbonatite

The Al Uyaynah carbonatites are divided on the basis of the chemical compositions

and grain size into two common magma types: calcio-carbonatite and ferro-carbonatite

(Figure 3.3). These magma types are found in a majority of carbonatites worldwide [5]. The

carbonatite complex is comprised of many contiguous layers; and each layer thick is roughly

between 10 to 15 meters. These layers contain lapilli, which requires eruption within a sub-

aerial environment [5]. Calcite is the major mineral phase in the calcio-carbonatites, but the

calcio-carbonatites also include a significant amount of apatite, as well as opaque minerals

such as magnetite. The typical composition of the lapilli is 40% calcite, 20-30% apatite, and

20-40% opaque minerals [5]. Based on the current whole-rock geochemical data, ferro-

carbonatites exhibit larger abundances of silicate minerals compared to calcio-carbonatites.

Representative ferro-carbonatite minerals are allanite (Ce), amphibole (alkali), and

occasionally strontian barite. Allanite typically occurs as crystal accumulation in ferro-

carbonatites [2]. These accumulations concentrate REEs, resulting in a heterogeneous

dispersion of trace elements within the rock [5].

3. Methods

3.1 ASTER data calibration

ASTER multispectral bands (VNIR-SWIR) that were utilized in this research were

acquired from the Land Processes Distributed Active Archive Center (LP DAAC) which is

based at the USGS Earth Resources Observation and Science (EROS) Center in Sioux Falls,

72

South Dakota. ASTER data was acquired of the Al-Uyaynah area on 02-05-2002. The data

was obtained as AST_07 Surface Reflectance and Crosstalk Corrected SWIR as Hierarchical

Data Format (HDF) product [25].

The ASTER AST_07 product is geo-referenced in Universal Transverse Mercator

(UTM) map projection arrangement and World Geodetic System datum (WGS-84) [25]. It

already includes atmospherically corrected data of Level-1B. The surface reflectance of

VNIR was produced by utilizing the 3 VNIR bands (0.52 - 0.86 µm). The SWIR surface

brightness is created by applying a crosstalk correction algorithm, then utilizing 6 SWIR

bands between 1.60 - 2.43 µm of a Level-1B of ASTER scene.

3.2 ASTER data analysis

3.2.1 Dimensionality reduction

The Minimum Noise Fraction (MNF) method is a linear transformation utilized to

differentiate noise from data and identify inherent spectral information dimensionally [13].

Each of the 9 measured ASTER bands (0.52 - 0.86 µm) were incorporated for processing in

the MNF transformation. MNF was applied to the scene in order to recognize and evaluate

which bands contained data that was coherent, and which had noise. This was achieved by

examining the MNF gray bands along with eigenvalues diagram (Figure 3.4 and 5). (see the

result section).

73

Figure 3.4. A MNF eigenvalue plot for study area (Al-Uyaynah). All MNF bands have eigenvalues values

greater than 1. The lowest band (9) has eigenvalue equal to 2.53, which means it contains coherent data.

However, the first 5 bands contain more spatially coherent information than the other bands.

74

Figure 3.5 MNF images of ASTER VNIR-SWIR bands for study area. Images 1-9, display a single gray

band, while image 10 shows a combination of first three MNF bands (1, 2 and 3 as RGB) for comparison

75

3.2.2 Endmember extraction

The study image has present within it many different rock units. Each unit exhibits

different spectral values, which must be de-convoluted. Finding and extracting the

representative endmembers is an important step that must occur prior to the image

segmentation processing of our study area. In this research, the endmembers were determined

and extracted from ASTER data by utilizing the Pixel Purity Index (PPI) tool of ENVI

software. The Pixel Purity Index (PPI) represents the intensity of pure pixels within an image

[13, 26]. The PPI is often determined by projecting the n-D (n-dimensional) scattergraph on

an indiscriminate unit vector [13, 26]. The PPI functions were applied on the MNF images

with 5,000 iterations, and then the results were visualized using n-D visualizer (n-D scatter

plots). Pixels located on the corners of the unit vector were assigned to be the spectral

reflectance end-members (Figure 3.6). After that, the extracted end-members were utilized

to classify the ASTER image and map the carbonatites (see the result section).

3.2.3 Defining the regions of interest (ROIs)

Regions of Interest (ROIs) are defined as potential targets of a raster that were

identified based on a geological map of the area and field information. For this project, we

focus on areas of carbonatite, and examine the correlation between previously identified

exposure and predicted occurrences in our post classification accuracy assessment [13]. ROIs

were identified in this study based on the 1:50 000 lithological map [1], and ground-truth

76

Figure 3.6. Show a 3-D Scatter plot of the end-members data cloud (image bands 1-3 were used as plot axes).

The identical rock units were characterized by the same color.

77

locations or rock units that were collected during our field trip. The ROIs were defined using

ENVI 5.1 software ROI tool and were based on ASTER image coordinates. The defined ROIs

were utilized to produce a random sampling of pixels (Figure 3.7). These random sampling

of pixels were utilized to evaluate the mapping outcomes of the BEC, SID, SAM, and SVM

algorithms (See section 5.3. Classification results and accuracy assessment).

3.3 Classification methods

The image categorization method can be employed on the basis of spectral analysis

approach and statistical classification techniques that summarize relevant pixels in

meaningful categories [27, 28]. For image examination and analysis, the spectral match base

methods are more feasible given that the rocks and minerals usually tend to have distinct

spectral absorption characteristics[27, 29]. Quite a few image categorization techniques rely

on the spectral method of analysis and have been developed for charting out geological maps

[30]. They include the Spectral Information Divergence (SID) algorithm, the Binary

Encoding Classification (BEC), the Spectral Angle Mapper (SAM), and Support Vector

Machines (SVM). In this study, these approaches were examined to determine the most

robust method to discover and map carbonatite and other relevant compositions within the

area concerned. Classification results were compared, so as to assess these algorithms, with

the data from the field and geological maps.

78

Figure 3.7. Shows a random sampling of pixels that were generated from the defined ROIs. These

random pixels were used as a reference to assess the mapping outcomes of the BEC, SID, SAM, and

SVM methods.

79

3.3.1 The Binary Encoding Classification (BEC)

The Binary Encoding Classification (BEC) is a method that encodes the endmember

spectra information and image data into binary numbers and is dependent on whether the

band values fall below or above the mean of spectrum [31]. This algorithm analyzes and

compares or matches every encoded endmember through the encoded image spectra and then

creates a category image [13, 31].

In order to make a lithological map in the present study, the binary encoding match

algorithm was employed due to the many advantages it offers including: 1) it is a relatively

simple algorithm [31]; 2) it is quite efficient for classifying minerals/elements with

differentiated absorption characteristics [31, 32]; 3) it is also useful to measure the process

of extraction since it has the sensitivity to band positions and is unaffected by albedo

(brightness) difference [32]. Consequently, this method might be useful for mapping of the

complex nature of our study area using medium special resolution (ASTER data), (See the

section 5.2.7 for results of applying this method).

3.3.2 The Spectral Information Divergence (SID)

The Spectral Information Divergence (SID) technique, which is derived from the idea

of the theory of divergence in information, is applied to explain the spectrum statistics [33].

This method employs a divergence measure to assign pixels into endmember spectra [13,

80

34]. The likeness between the spectra among pair pixels is measured by examining the

amount of difference of probabilistic behaviors amongst their spectra [35]. The smaller the

divergence is between pixel and endmember, the greater the probability that the pixels of

endmember and image spectra are comparable [13, 34]. The pixels with measurements that

are larger in relation to the maximum divergence threshold are not categorized [13].

For the study of carbonatite, the SID algorithm has potential utility as it is able to

measure the spectral variance of an individually mixed pixel from a probabilistic viewpoint

[35]. This feature is useful given that each pixel of our research area image encompasses

mixed rock units due to the medium resolution nature of the dataset and the size of the

outcrops. The SID algorithm also has the benefit of finding the spectral similarity between

pixels and endmembers and exhibits effective separability among of the mineral end-

members [36]. Therefore, this method is particularly useful for mapping of the complex

geology of our study area using ASTER datasets.

3.3.3 Spectral Angle Mapper (SAM)

The SAM algorithm is a physical technique used to identify an image focused on

fitting and comparing the angle in the space between the pixel vector to an end-member

image spectral vector in the n-dimensional space [13, 26, 37]. The smaller the angles between

the endmember and pixel, the more of a match there is with the endmember spectrum; where

81

Figure 3.8. Shows how the SAM method classifies the image. The smaller the angles between the

endmember and pixel, the more of a match there exists with the endmember spectrum. Where pixels surpass

the defined angle threshold, they are not categorized. The figure was modified from [37].

82

pixels exceed the defined rock unit endmemeber maximum angle threshold are not

categorized (Figure 3.8) [13].

In our region, the complicated topography and geology make distinct endmembers

less pronounced. The SAM algorithm has potential to resolve the distinct endmembers over

other more typically used spectral-based methods because: 1) the SAM technique is an

efficient technique for finding and outlining the likeness between the image spectra and

reference spectra [13, 37]; 2) SAM is

not impacted by shading which can enhance or isolate the rock units reflectance features [38],

and 3) SAM does not require any presumptions with regard to the statistical- distributions of

the input information in carrying out segmentation as is common in parametric procedures

[39].

3.3.4 Support Vector Machines (SVM)

Support Vector Machines (SVM) is a supervised classification approach produced

from the statistical learning theory that frequently produces efficient classification outcomes

from noisy and complicated data [34]. SVM dissociates the images units using a decision

surface (optimal hyperplane) that increase or maximize the border among the groups of

classes (support vectors) (Figure. 3.9) [13, 34].

83

Figure 3.9. Linear Support Vector Machine illustration. The SVM training algorithm is used to find a

hyperplane that divides the image pixels to a separate number of classes. Modified from [41, 42].

84

In this research, we have utilized the SVM method as it has the following strengths:

1) The method has the capacity to utilize a limited number of training data sets to regularly

deliver higher characterization precision than the conventional methods [40]; 2) The

learning-procedure central to the method (structural risk minimization), offers a good

classification without needing prior knowledge, even with very high dimension of the input

space [41]; 3) SVM is a non‐parametric

classification and commonly used in studies where noisy and complex data are present [42];

4) The SVM learning method discovers the decision boundary or hyperplane between the

image data with iterative process to separate the likely groups, based on the defined

endmembers, and then maximizes the distance between each group from the hyperplane

(Figure. 3.9) [43, 44]. The hyperplane option that is available in the SVM method was

employed to help to decrease the chances of misclassification in our study area; 5) SVMs

have been shown to be generally robust since as they have control parameters options [42].

3.4 Post classification assessments

A comparison and accuracy calculation was made to allow a better understating of the

different products' relative performance in the identification of lithologies for our ASTER

scenes. The evaluation of the performance of BEC, SID, SAM, and SVM is used in this study.

The computing of kappa coefficients and overall accuracies were calculated using a

85

confusion-matrix [13, 45]. The confusion-matrix (error matrix) approach is most extensively

used to characterize the classification accuracy of digital mapping of remote sensing data;

the error matrix is appropriate for many accuracy assessments [45]. Many scholars have

suggested utilizing the “error matrix” method in reviewing the accuracy of the mapping

product, and it is currently counted as a standard method for accuracy measurement (e.g. [46-

48]). In order to enhance measurements of accuracy, the kappa coefficient statistical tool is

also frequently applied[49]. However, there remains some disagreement as to the utility of

standard accuracy estimation measurements like the kappa coefficient for the calculation of

the accuracy of remote sensing classified images [50, 51]. Specifically, much of the remote

sensing information is influenced by diversified pixels, which may confound simple

statistical tools (e.g.[50]). Nevertheless, it remains valuable as an indicator of classification

result accuracy. Specifically, the kappa coefficient is still considered to be the most

appropriate method for rank-ordering of the mapping results accuracy’s agreement [52].

One of the most useful ways to utilize Kappa for mapping results assessment and

comparison was provided by Landis and others [52, 53]. They have categorized the different

ranges of kappa coefficient values based on based on the agreement degree [52]. They

suggested that the values greater than roughly 0.75-0.85 indicate a very good agreement;

values greater than 0.85-0.99 indicate excellent agreement, and values of close or equal to

1.0 would be perfect agreement. In contrast, lesser levels of agreement were classified as:

good if the kappa coefficient values were between 0.4-0.55; fair agreement between 0.40-

0.55 values; and values of 0.4-0.05 or less indicate poor to no agreement as illustrated in

86

Table.2 [52, 53]. It is worth mentioning that the accuracy, meaning, and measurement of

classification may depend ultimately on the demands of an individual project [47, 54]. In this

research, we used the accuracy calculation for evaluating and ranking the segmentation

results from BEC, SID, SAM, and SVM algorithms (See section 6. Classification results and

accuracy assessment).

4. Field surveys and ASD measurements

Rock samples and location information were gathered during field surveys in

December 2010. The weathered and fresh surfaces of the returned samples were examined

and analyzed utilizing an Analytical Spectral Device (ASD) spectroradiometer (Fig. 3.9). All

reflectance spectral measurements were made in the laboratory utilizing an ASD FieldSpec-

3 Hi-Res Spectroradiometer, which is located within the Department of Environmental

Sciences at the main campus of the Univesity of Toledo. A 1000W 3200K Quartz Halogen

lamp provided illumination. A 10-degree foreoptic was employed. Reflectance spectra were

obtained for both unweathered and weathered surfaces for every sample. All spectra were

normalized to a Spectrlon lab standard with well characterized spectral properties. All ASD-

Laboratory spectra were resampled to the ASTER system (VNIR-SWIR) using spectral

resampling tool available in ENVI software [13].

87

5. Results

5.1 Noise reduction and endmembers extraction

ASTER bands covering the VNIR and SWIR spectral range (0.433–2.95 µm) were

selected for the MNF transformation procedure. The outcomes of this transformation

contained eigenvalues and eigenimages (MNF images), which were used to evaluate the data

dimensionality by comparing MNF pictures with MNF eigenvalues (Figure 3.4). Bands with

higher eigenvalues (greater than 1) comprise the main share of inherent information, and

bands with lower eigenvalues (≤ 1) contain noise [13]. The eigenvalues for all MNF bands

are above 11.23. However, more spatially coherent data were present in the first three bands

(2 to 3) as compared the rest of the bands (4-7), as illustrated in Figures 3.5.

In order to determine endmembers, the MNF images containing coherent data (1-9

bands) were analyzed utilizing the PPI technique with 20,000 iterations. These end-members

were then visually examined using an n-D visualizer (n-D scatter plots). Pixels at the corners

of each dataset were then selected as class types using the region-of-interest (ROI) tool. After

the purist pixels had been defined by using n-D scatter-graphs, the distinct pixels were

marked and recognized as regions of interest (ROI) with the ASTER Image. The calculations

for mean-spectra for ROI classes were extracted and were utilized as endmembers for further

image classification. Additionally, a supported field based ground-truth (ROI’s) mean-

spectra were used along with image endmember spectra in classification and rock units

mapping.

88

Figure 3.11. The resampled laboratory reflectance spectra (ASD) of ferro-carbonatite and calcio-

carbonatite.

Figure 3.10. Laboratory reflectance spectra (ASD), showing REE absorption near 0.589, 0.567,

and 0.56 µm and (CO3, MgO) absorption near 2.32-2.35µm, and near 1.6 µm for calcite absorption.

89

Figure 3.12. A zoom-in of the spectral region 0.5-0.6 µm of laboratory reflectance spectra (ASD),

showing REE absorption near 0.589, 0.567, and 0.56 µm (left), and a zoom-in of the spectral region

0.2-2.4 µm of laboratory reflectance spectra (ASD) showing (CO3, MgO) absorption near 2.32-

2.35µm, and near 1.6 µm for calcite absorption.

90

Figure 3.13. ASTER image endmemebers spectra of calcio-carbonatite and ferro-carbonatite; in

VNIR-SWIR bands.

91

5.2 Endmembers' spectral characteristics

To unravel the endmember spectra, we have evaluated the possible causes for the

distinct absorbencies in the ferro- and calcio carbonatite. Broadly we have found that calcite,

iron, REE, and silica-based minerals dominate the spectra.

5.2.1 Calcite

Calcite is the major mineral phase in calcio-carbonatite, constituting about 40% of its

lapilli composition. While less abundant, significant volumes of calcite are also present in

ferro-carbonatites. The calcio-carbonatite and ferro-carbonatite image endmembers show a

clear calcite (CO32- ion) absorption feature at 2.336 µm (Fig. 3.12). This is consistent with

the ASD spectra of both carbonatite types, which also show a strong absorption of calcite at

2.336 µm (Fig. 3.10- 3.12).

Some distinction between the two rock types is evident due to the greater percentage

of calcite within the calcio-carbonatites, which manifests as more distinct absorption peak at

2.336 µm in both ASD and image end-members in comparison to ferro-carbonatites (Fig.

3.10, and 3.12).

92

5.2.2 Alkali amphibole minerals

The alkali amphibole group of minerals are a major constituent of the ferro-

carbonatites, but only a secondary mineral in calcio-carbonatite [5]. The hydroxyl ion

associated with the hydrosilicates, Mg-OH and Fe-OH, of alkali amphibole [55, 56] generate

distinct absorption features at 1.4 µm and 2.2-2.3 µm [8]. The ferro-carbonatites image

endmember displays hydroxyl ion (OH) absorption at 2.26 µm, caused by the Mg-OH and

Fe-OH components of alkali amphibole minerals. The ferro-carbonatites ASD laboratory

spectra is consistent with the image endmember absorptions. However, it also displays a

greater number of OH absorption features at the ranges 1.4, and from 2 to 2.226 µm of the

spectrum (Fig. 3.10).

In contrast, the alkali amphibole forms as only a secondary mineral within the calcio-

carbonatite. Some minor absorption near 1.4 µm of ASD spectrum was observed (Figure.

3.10). However, this absorption is not detected in the ASTER image calcio-carbonatite

endmember (Figure 3.13), because the atmospheric (H2O vapor) absorption prevents the use

of these band ranges (0.9-1.6 µm) for reliable mineral identification.

5.2.3 Iron minerals

Ferrous and ferric ions (Fe2+, Fe3+), are present in iron oxide minerals in both of ferro-

carbonatite and calcio-carbonatites outcrops. This is quite evident in the field as a rust-

weathered surface, as rough outcrop with a crust of iron oxides. Iron minerals, magnetite and

93

hematite, are among the major minerals forming the ferro-carbonatite matrix, however, iron

minerals are secondary minerals in the calcio-carbonatite [5]. Iron minerals diagnostic

absorption features were recorded by many scholars, (e.g., [17]), within the very-near

infrared (VNIR) ranges. In our study area, the ferro-carbonatite and calcio-carbonatite spectra

show that the iron oxides were recorded by ASD and ASTER instruments for the weathered

outcrops in both of ferro/calcio-carbonatites near 0.804 µm, for both image end members and

ASD spectra (Figure 3.10, and 3.12).

Ferro-carbonatites have more iron minerals (magnetite and hematite) than calcio-

carbonatite. These iron minerals caused additional absorption in the area between 0.661-

0.804 µm of the ferro-carbonatite image end member and ASD spectra, and weak Fe3+

absorptions at 2.26 µm, and this may facilitate a differentiation between the two carbonatite

spectra.

5.2.4 REE minerals

Rare earth elements (REE) are found in great abundance in carbonatite lavas,

frequently forming distinctive mineralogies. Within our study area, ferro-carbonatite has a

significant abundance of rare earth element bearing minerals, while calcio-carbonatite

contains significantly lower concentrations [5]. REE spectral absorption features have

previously been reported in the USGS digital spectral library within the visible region [17].

Our ferro-carbonatite ASD spectra show REEs absorption features in the area between 0.45

94

- 0.56 µm (Fig. 3.12). However, the ASTER image endmembers did not show REEs

absorption features due to the ability detection limits of medium spectral and special

resolution sensor (ASTER).

In contrast to ferro-carbonatites, neither calcio-carbonatite the image endmember nor

ASD spectra of calcio-carbonatites exhibited REEs absorption characteristics. This is the

result of likely low REE abundances in calcio-carbonatite lavas.

5.3 Classification and accuracy assessment

Evaluation and accuracy calculations permit us to develop a better understating of the

various mapping products performances in the classification of

lithologies from ASTER scenes. The determination of kappa coefficients and overall

accuracies were calculated using a confusion-matrix method [13, 45]. The comparison of the

confusion matrix overall accuracy information and the kappa coefficient results for the

analysis of four different mapping methods are illustrated in table 3.1 and figure 3.20. The

evaluation of the performance of each method (BEC, SID, SAM, and SVM) is addressed in

the sub-sections below.

95

Figure 3.14. The image endmember spectra were extracted from Landsat OLI image and are

utilized as endmembers for further image classification.

96

5.3.1 Binary Encoding Classification (BEC)

The first method we tested was the Binary Encoding Classification (BEC). The BEC

algorithm encodes endmember spectra and image data into zeros and ones and then compares

each encoded data point to create a map [13, 31]. The overall accuracy of the BEC map is 39

%. The calculated kappa coefficient, which measures classification performance relative to

chance, is 0.31, (Fig. 3.15). These results are indicative of ‘No agreement’ in the

classification standard category (Table 3.1, and Fig. 3.20) [52, 57].

5.3.2 Spectral Information Divergence (SID)

The second method utilized was the Spectral Information Divergence (SID) approach.

The SID map shows a fair discrimination among ferro/calcio-carbonatite and other the rock

units (Fig. 3.16 and 3.20). The accuracy assessment of the SID shows overall accuracy 52%.

The kappa coefficient is 0.47, which falls into the ‘fair category [52, 57].

5.3.3 Spectral Angle Mapper (SAM)

The Spectral Angle Mapper (SAM) classification was employed in the calculation of

the spectral matches between endmembers and image spectra [21-23]. SAM achieves the

second best results (Figure 3.17 and 3.20) with an overall accuracy of 61 %. The calculated

Kappa Coefficient is 0.56, placing this method into the ‘good’ agreement category of the

kappa coefficient statistic.

97

5.3.4 Support Vector Machine (SVM)

The Support Vector Machine (SVM) digital-mapping results shown in Figure 18

achieved the best results, with an overall accuracy of 86.4% (Figure 3.18). The calculated

kappa coefficient is 0.85. This is recognized as ‘excellent’ in the classification standard

category (Table 3.1, and Fig. 3.20) [52, 57].

6. Discussion

Our experimental approach of using four different methods (BEC, SID, SAM, and

SVM) to map and differentiate carbonatites from other rocks unit highlighted significant

performance differences among the algorithms (see the table’s 3.2-3.5 in appendix). The

observed variable results likely reflect the following issues when interpreting data from this

region: 1) the variable topography of the study area leads to the data being complicated by

shadows or different illumination factors in the same unit or between rock units; 2) the

minerals' similarity in composition within rock formations, e.g., the existence of silicate

minerals in many units, contributes to the complexity of some algorithms' performance

during the analysis process; 3) The image 30 m2 pixel (medium resolution) results in quite

mixed pixels with multiple rock units in each pixel; Despite these practical difficulties, some

algorithms performed well, while others did not. Below we outline the particular

characteristics of each algorithm which may have impacted its performance in this instance.

98

Figure 3.15. The binary encoding classification (BEC) map shows very poor discrimination


accuracy is 39%, the Kappa Coefficient is 0.31, this recognized as "No agreement" in the

classification standard category.

99

Figure 3.16. The Spectral Information Divergence (SID) map shows fair differentiation between

rock units. The accuracy assessment of the SID shows an overall accuracy with 52%; the Kappa

Coefficient is 0.47, which is known as “fair results" in the category.

100

Figure 3.17. The Spectral Angle Mapper (SAM) classification performs the second best result

with an Overall accuracy of 61%. The calculated Kappa Coefficient is 0.56 placing into the good

agreement category of the Kappa Coefficient statistic.

101

Figure 3.18. The Support Vector Machine (SVM) digital-mapping shows the best result with an

overall accuracy of 86.4%. The calculated Kappa Coefficient is 0.85 placing into the excellent

agreement category of the Kappa Coefficient statistic.

102

Figure 3.19. Geological map of Al-Uyaynah area showing the mapped area of extrusive

carbonatites (modified from [1].

103

Lower Bound Degree of Agreement Upper Bound

<0.05 No 0.05

0.05 Very Poor 0.20

0.20 Poor 0.40

0.40 Fair 0.55

0.55 Good 0.70

0.70 Very Good 0.85

0.85 Excellent 0.99

0.99 Perfect 1.00

Table 3.1. The values of threshold that used for dividing the results of statistical kappa coefficient

degrees of agreement. From [52, 57].

Figure 3.20. A comparing of confusion matrix over all accuracy, and kappa coefficient results.

The binary encoding classification (BEC) overall accuracy is 43.59%; the calculated Kappa

Coefficient is 0.32. This recognized as No agreement ‘not even poor results’ in the classification

standard category. The Spectral Information Divergence (SID) map shows a good discrimination


accuracy is 68.3%; the Kappa Coefficient is 0.58, which is known as "good results" in the

category. SAM performs the second best result with an Overall accuracy of 81.54%. The

calculated Kappa Coefficient is a mere 0.75 placing into the very good agreement category of the

Kappa Coefficient statistic. The Support Vector Machine (SVM) digital-mapping results with an

overall accuracy of 97.58%. The calculated Kappa Coefficient is 0.97. This recognized as

excellent in the classification standard category.

BIC SID SAM SVM

Kappa Coefficient 0.3062 0.4605 0.5619 0.8464

Overall Accuracy 38.67% 51.74% 61.02% 86.32%

0.00%

10.00%

20.00%

30.00%

40.00%

50.00%

60.00%

70.00%

80.00%

90.00%

100.00%

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

OV

ER

AL

L A

CC

UR

AC

Y

KA

PP

A C

OE

FF

ICIE

NT

Kappa Coefficient Overall Accuracy

104

The BEC algorithm achieved poor results since it was designed to encode the

information of image pixels and end-member’s spectra into binary numbers. Encoding

spectrum, to 0 or 1, has a drawback in lithological mapping since: 1) the study area has a

complicated geology. Encoding causes a loss of some of the valuable details for both the

endmembers spectral characteristics and image information; 2) The BEC has classified the

image based on the winning class based on whether the band values fall beneath or over the

class mean [31]. In our study, many rock units exhibit an abundance of silicate minerals, and

the similarity in these minerals make the rock units hard to distinguish using this method.

The SID algorithm shows a fair result, but it did not generate a better map than SAM

and SVM methods for many reasons: 1) SID analyzes the spectra as random or irregularly

variable and then measures the divergence between spectra [35]. Our research area has: 1)

complicated lithological features, consequently this stochastic approach did not consider the

spatial arrangement of some rock units and thus produced an abnormal map; 2) SID is less

sensitive to the shadow factors in the spectrum as compared with the SAM method [58]. This

decreases the spectral values of outcrops in the shadowy areas, and produces a less accurate

mapping result; 3) In our multispectral image, a pixel is commonly a mixture of many of

rocks with similar spectral features, SID is unable to resolve these complex pixels effectively.

The SAM algorithm achieved a ‘good’ mapping result for the following reasons: 1)

This method is effective at grouping image pixels, because the angle is independent between

the endmemeber image pixels [37]; 2) it also resistant to the impact of shading and enhances

105

the rock's spectra by isolating its reflectance spectra from the shadow effects [38], and 3)

SAM does not require any statistical distribution assumptions of the data in order to create a

classification - as is common in other parametric procedures [39]; 4) the only drawback of

this technique is the problem of spectral mixtures in pixels as described above. SVM method

was judged the best technique and achieved the ‘excellent’ result since: 1) the SVM is a linear

method that discovers hyperplane, allowing the method to separate the likely rock groupings.

By increasing the distance of units from the hyperplane, the hyperplane agent reduces the

chance of misclassification [43]; 2) SVM was designed based on a statistical learning

machine (theory) that frequently yields efficient classification outcomes from noisy and

complicated data [41]; 3) because of its capacity to operate with a small number of training

classes, it is capable of performing at a higher characterization precision than the other

methods [40].

The calcio/ferro-carbonatites n-D spectra show that the calcio/ferro-carbonatites

distribution in the central region of the study area, measured from north to south, closely

matched both field ground truth ROIs reference geological maps. SVM and SAM

classification maps show calcio/ferro-carbonatites distribution in the central region of the

study area, measured from north to south. SVM was the most successful in distinguishing

between major rock units, particularly in the discrimination of calcio-carbonatites (98.96%

of class accuracy), and ferro-carbonatites with 91.01% accuracy (Figure 3.24, and Table 3.5).

SAM was able to discriminate calcio- carbonatites with 57.29% accuracy, and correctly

identified 87.64% of ferro-carbonatites (Figure 3.23, and Table 3.4). Calcio/ferro-

106

carbonatites were distinguishable by both SAM and SVM algorithms, but the SVM method

gave the best result after comparing SVM to field ground truth regions of interest. However,

the classification error in calcio-carbonatites and in ferro- carbonatites SVM map were found

to be a result of three factors. Firstly, it is possible for some carbonatites to remain undetected

due to the fluctuation in the sizes of carbonatite bodies. Secondly, it is possible for sub-pixel

contamination to take place because of the varying types of rock present in the ASTER image

within a 30 square meter pixel size. Lastly, there is possible imprecision caused by the

spectral similarity between metavolcanic rocks, carbonatite units, and the alkaline volcanic

formation in the neighboring geographic area. The calcio and ferro-carbonatite accuracy

results may also be impacted by the low numbers of pixels (<100 pixels) that were selected

at random by the ENVI software when evaluating the mapping outcomes.

The alkali volcanic unit was mapped by employing SVM and SAM algorithms, in the

eastern and western regions of carbonatite units, which is dominated by deformed basaltic

alkali volcanic rocks (sub/green schist facies). The SVM algorithm mapped this unit with

72.48% accuracy, while the SAM method yielded 36.70% accuracy. The 27.52% error in the

SVM map was due to the likeness of the absorption feature of silicate in alkali basalts and

ferro-carbonatites. These resemblances generated some confusion for the SVM method

during the analysis and created an overlap between ferro-carbonatites and the alkali volcanic

unit in the southern area of our study.

107

The SVM and SAM maps of the metavolcanic unit correspond closely with the

previously mapped metavolcanic unit in the western part of the geologic map. The SVM class

accuracy for the metavolcanic unit is 78.26%, though it much poorer with SAM at 40.22%.

The error likely relates to the composition of the metavolcanic units, which are rich in

silicates, as are the alkali volcanic rocks and red-chert. Therefore, these spectral similarities

caused overlap between rock units in contact areas; especially, in the eastern portion of the

study area where accuracy is less robust.

The ophiolite complex, is well characterized using the SVM and SAM methods. The

image results matched the ROI’s and available geological maps, classify ophiolite with

99.43% accuracy (SVM), and 94.29% with the SAM method. The ophiolite mapped with

high accuracy using both SVM and SAM because there is a large non-distributed mass of the

ophiolite unit in study area image (about 11.7 Km2), which resulted in enhanced classification

performance (Figure 3.24, and Table 3.5), and (Figure 3.23, and Table 3.4).

Red chert as indicated in geological maps is in the regions between alkaline

metavolcanic units to the west and the ophiolite unit to the eastern side of the map, and

extends in north to south orientation. The red chert unit was mapped with SVM at 78.74%

accuracy but only 47.24% accuracy with the SAM classification scheme. However, we

noticed that the red-chert unit in the south region of the study area was categorized as

metavolcanic by the SVM algorithm, despite the overall low error in terms of absolute

percentage. The red chert was badly categorized using the SAM method, as it relies on the

108

matching of the spectral features; both the red-chert and metavolcanic units have a high

percentage of silicate minerals, leading to errors in this technique.

7. Testing SVM as a carbonatite detection tool - performance at an alternate locality

In the above mapping experiments, we have found that the SVM algorithm achieved

the best results among all tested methods. To verify the utility of the SVM technique for

carbonatite identification and mapping, we examined an alternate area which exhibits

carbonatites. This area has similar tectonic and petrogenetic environment. The Hatta

carbonatite area is located approximately 60 km to the south of the main study area (Figure

3.26). This area is located on the west side of Semail Ophiolite Complex, within the Hatta

transform fault zone, UAE. The extrusive carbonatite in Hatta zone was discovered within

the deep sea metasedimentary unit and is comprised of lenses and layers [2]. The layers

average thickness range between 2-50m (Figure 3.25), and the weathered outcrop of these

layers are colored with dark brown from the iron oxides [2]. These layers’ exhibit calcite and

silicate (quartz) veins. In our primary study area (Al-Uyaynah) there are two types of

carbonatite (ferro/calcio-carbonatite) present. Despite the Hatta zone sharing the same

petrogenetic history and major geochemical background as Al-Uyaynah carbonatite, in the

Hatta zone, there is only one kind of carbonatite, which was found to be similar to ferro-

carbonatite [2]. The Hatta ferro-carbonatite lapilli in this area consist of carbonate, apatite,

monazite, allanite, and titanite as major minerals. The minor minerals are alkali amphibole,

chlorite, and quartz [2].

109

We used ASTER (VNIR-SWIR) bands to analyze and map Hatta carbonatites. The

data was obtained from Land Processes Distributed Active Archive Center (LP DAAC) as

AST_07 as surface reflectance and crosstalk corrected SWIR bands as Hierarchical Data

Format (HDF) product [25]. The data were already geo-referenced in UTM projection system

and World Geodetic System datum (WGS-84) [25].

We followed the same post classification procedures that were done over Al-Uyaynah

area as described above. The Minimum Noise Fraction (MNF) method was used to the image

nine bands in order to understand our data, calculate, and separate the coherent information

from the bands that had noise. The MNF bands along with eigenvalues plot diagram were

inspected, and it showed that we can use all nine band in the analysis and classification

because it contains inherent information. The Endmembers were extracted by using the Pixel

Purity Index (PPI) tool in order extract of the represented endmembers of rock units. The iron

oxides in the carbonatites, and carbonate minerals, were key characteristics used in

discovering the carbonatite spectra during pre-classification passes. Based on these end-

members we performed the SVM classification. The Support Vector Machine (SVM) method

was applied to the Hatta image and achieved ‘excellent’ mapping result, illustrated in Figure

3.27, with an overall accuracy of 99.85%. The calculated Kappa Coefficient is 0.99. The

carbonatite was mapped with this method with about 86% accuracy. The SVM miss-mapped

carbonatite with 14% error due to the similarity in minerals between the meta-sediments and

ferro-carbonatite.

110

8. Conclusion

This research presents a review of ASTER multispectral information analysis and

interpretation in visible/near-infrared (VNIR), and Short Wave Infrared (SWIR) ranges for

lithological mapping of carbonatites and surrounding rock units. ASTER is one of the best

multispectral-remote sensing sensors for mineral mapping. A medium spatial resolution (15-

30m) of ASTER images provides enough spectral ranges to detect carbonatite complexes,

which can be relatively narrow.

In this study, we have successfully distinguished and mapped eight rock units using

ASTER data, and generated acceptable separability between rock formations. The best

differentiation between end-members was obtained through use of the SVM method, yielding

86.3% (excellent) overall agreement with the field ground truth ROIs and the existing

published geological map. While useful, the SAM algorithm provided a map with 61.03%

overall agreement (Good) with the field’s ROI data and the geological map.

Our primary targets are the extrusive ferro- and calcio-carbonatites rocks - these units

were well classified using these techniques. Ferro-carbonatite rocks were mapped with

91.01% accuracy by SVM, but are much poorer resolved using SAM (87.64%). The calcio-

carbonatite endmember is similarly well resolved, 98.96% accuracy using SVM, and 57.29%

of class accuracy with the SAM. While excellent classifications, some degree of error is

noted:

111

(1) Mapping of ferro-carbonatite was more difficult at the southern region using both

methods (SVM and SAM), as in this region deformed structures are present resulting in

comparatively thin/ narrow rock units (a few meters in width). This problem is particularly

acute near to the mélange (alkali volcanic) unit. The mixed appearance of the rock units’

spectral likeness can contribute to the problematic calculations within the SVM and SAM

algorithms and results in a non-recognition of sub-pixel spectral contamination.

(2) Complexity in distinguishing between ferro- and calcio-carbonatite endmembers

results from the similar richness in iron oxide minerals in weathered surface layers of these

rocks, causing a marked similarity in spectra. This similarity in the appearance of altered

surfaces created some mis-classification in the SVM and SAM classifications in certain areas

of the upper northern portion of the carbonatite complex. Accordingly, weathered/altered

calcio-carbonatite might be mapped as ferro-carbonatite because of the shared iron oxide and

hydroxide spectral characteristics.

(3) Our techniques identified carbonatite pixels in regions where the geologic map

does not indicate the presence of this unit. However, given the difficulties in mapping

carbonaite in the field, it may have escaped detection in prior maps. Indeed, smaller

occurrences of carbonatite are geologically feasible in any of the units below the Samail

ophiolite. Thus this error may in fact result from a more accurate determination of carbonaite

occurrence in the mixed and complex units of this area.

112

Other geologic units in the region that are rich in silicates and hydroxyl minerals

(alkali volcanic mélange, metavolcanic and chert formations) share similar vibration energy

features, making them difficult to robustly classify. However, despite the similar

characteristics, the SVM separated these units with ‘excellent’ degrees of accuracy while

SAM separated these units with ‘good’ levels of precision. Given such units may commonly

be found in the vicinity of carbonanites in other environments (in particular the alkali

volcanic units), this separation is important for the broader application of the technique

presented herein.

Through comparison of field data and the geological map of the Al-Uyaynah

extrusive carbonatite, the moderate spatial resolution results derived using SVM, SAM, SID,

and BEC algorithms, the following findings have been reached: ‘excellent’ results were

acquired with SVM method and a ‘good’ results have been obtained using SAM algorithm

regardless of the spatial resolution restrictions of ASTER data, along with the reality

endmembers with complicated sub-pixel mixing between irregular rock types. The

categorization map created using SVM from ASTER image; illustrates that this process

SVM) provides an efficient method for carbonatite rocks/mineral exploration and geologic

mapping of these type of rock units in general.

113

APPENDIX

114

Figure 3.21. A Comparison between the rocks unit’s distribution that classified correctly, in percent, of BEC map based on the results of the confusion matrix

accuracy assessment. BEC method gave, overall, the poorest result after comparing BEC to the field ground truth regions.

Chert

formation

Metavolcanic

formation 1

Schist

formationWadi 1

Alkaline

volcanic rocks

Calcio-

carbonatite

Ferro-

carbonatite

Metavolcanic

formation 2

Ophiolite

complex

Ophiolite complex 0.79 9.78 0 0 0 0 0 13.24 88

Metavolcanic formation 2 0 0 0 0 0 0 0 0 0

Ferro-carbonatite 0 0 0 0 0 0 0 0 0

Calcio-carbonatite 28.35 40.22 6.49 4.17 38.53 48.96 3.37 26.47 0

Alkaline volcanic rocks 8.66 32.61 2.6 0 50.46 44.79 95.51 57.35 0

Schist formation 14.17 6.52 63.64 38.89 0 0 0 0 2.86

Metavolcanic formation 1 0 0 0 0 0 0 0 0 0

Chert formation 29.13 9.78 10.39 15.28 5.5 6.25 1.12 1.47 9.14

0

10

20

30

40

50

60

70

80

90

100

BEC

Overall Accuracy = 38.6735%

Kappa Coefficient = 0.3062

115

Figure 3.22. A Comparison between the rocks unit’s distribution that classified correctly, in percent, of SID map based on the results of the confusion

matrix accuracy assessment. SID method gave, overall, a fair result after comparing SID to the field ground truth regions.

UnclassifiedChert

formation

Metavolcanic

formation 1

Schist

formation

Alkaline

volcanic rocks

Calcio-

carbonatite

Ferro-

carbonatite

Metavolcanic

formation 2

Ophiolite

complex

Ophiolite complex 0.57 0 7.43 1.14 0 0 0 1.14 77.14

Metavolcanic formation 2 0 0 23.53 0 8.82 2.94 20.59 36.76 2.94

Ferro-carbonatite 0 0 0 0 4.49 4.49 89.89 1.12 0

Calcio-carbonatite 2.08 11.46 0 0 10.42 57.29 18.75 0 0

Alkaline volcanic rocks 0.92 14.68 3.67 0 45.87 6.42 23.85 0.92 0

Schist formation 6.49 7.79 7.79 55.84 1.3 0 0 2.6 0

Metavolcanic formation 1 0 15.22 20.65 3.26 21.74 6.52 9.78 6.52 6.52

Chert formation 3.15 33.07 7.87 21.26 6.3 7.09 3.94 7.09 0.79

0

10

20

30

40

50

60

70

80

90

100

SID



116

Figure 3.23. A Comparison between the rocks units’ distribution that classified correctly, in percent, of SAM map based on the results of the confusion

matrix accuracy assessment. SAM method performs the second best result, overall good agreement, after comparing SAM to the field ground truth regions.

Chert formationMetavolcanic

formation 1Schist formation

Alkaline

volcanic rocks

Calcio-

carbonatite

Ferro-

carbonatite

Metavolcanic

formation 2

Ophiolite

complex

Ophiolite complex 0 0 0 0 0 0 0 94.29

Metavolcanic formation 2 8.82 20.59 0 0 0 16.18 54.41 0

Ferro-carbonatite 0 0 0 8.99 3.37 87.64 0 0

Calcio-carbonatite 19.79 1.04 0 10.42 57.29 11.46 0 0

Alkaline volcanic rocks 22.94 2.75 0.92 36.7 8.26 24.77 0.92 0

Schist formation 10.39 10.39 62.34 0 0 0 0 1.3

Metavolcanic formation 1 17.39 40.22 5.43 10.87 0 7.61 15.22 0

Chert formation 47.24 12.6 22.83 2.36 6.3 3.94 1.57 0

0

10

20

30

40

50

60

70

80

90

100

SAM



117

Figure 3.24. A Comparison between the rocks unit’s distribution that classified correctly, in percent, of SVM map based on the results of the confusion

matrix accuracy assessment. SVM method performs the second best result, overall an excellent agreement, after comparing SVM to the field ground truth

regions.

Chert formationMetavolcanic

formation 1Schist formation

Alkaline volcanic

rocks

Calcio-

carbonatiteFerro-carbonatite

Metavolcanic

formation 2

Ophiolite

complex

Ophiolite complex 0 0 0 0 0 0 0 99.43

Metavolcanic formation 2 0 4.41 0 0 0 1.47 92.65 0

Ferro-carbonatite 2.25 0 0 2.25 3.37 91.01 1.12 0

Calcio-carbonatite 1.04 0 0 0 98.96 0 0 0

Alkaline volcanic rocks 4.59 0.92 0 72.48 4.59 15.6 1.83 0

Schist formation 3.9 14.29 76.62 0 0 0 0 0

Metavolcanic formation 1 11.96 78.26 1.09 2.17 0 2.17 3.26 0

Chert formation 78.74 8.66 1.57 5.51 2.36 0.79 0.79 0

0

10

20

30

40

50

60

70

80

90

100

SVM



118

Figure 3.25. Shows the ferro-carbonatite layers that intruded into metasediments unit of the Hatta Zone.

Modified from [3]

119

Figure 3.26. shows the location and the geological map of the Hatta zone Modified from [2, 3]

120

Figure 3.27. The Support Vector Machine (SVM) digital-mapping of the Hatta carbonatite, shows the

best result with an overall accuracy of 99.85%. The calculated Kappa Coefficient is 0.99.

121

REFERENCES

122

REFERENCES

[1] K. M. Goodenough, E. Phillips, M. Styles, R. J. Thomas, A. Farrant, and S. Arkley,

Geology of the Khor Fakkan 1: 50 000 map sheet 50-3, United Arab Emirates: United Arab

Emirates, Ministry of Energy, Petroleum and Minerals Sector, Minerals Department, 2006.

[2] S. Nasir and R. Klemd, "New carbonatite occurrences along the Hatta transform fault zone

(northern Oman Mountains, United Arab Emirates)," Journal of African earth sciences,

vol. 27, pp. 3-10, 1998.

[3] D. Cooper, M. Ali, and M. Searle, "Structure of the northern Oman Mountains from the

Semail Ophiolite to the Foreland Basin," Geological Society, London, Special





"Extrusive Carbonatites from the Uyaynah Area, United Arab Emirates," J. Petrology, vol.

32, pp. 1143-1167, December 1, 1991 1991.


minerals and rocks; IV, Sulphides and sulphates," Modern Geology, vol. 3, pp. 1-14, 1971.

[7] F. v. d. Meer and S. M. d. Jong, Imaging spectrometry : basic principles and prospective

applications. Dordrecht ; Boston: Kluwer Academic Publishers, 2001.

[8] C. Elachi and J. J. Van Zyl, Introduction to the physics and techniques of remote sensing

vol. 28: John Wiley & Sons, 2006.

[9] R. N. Clark, T. V. King, M. Klejwa, G. A. Swayze, and N. Vergo, "High spectral resolution

reflectance spectroscopy of minerals," Journal of Geophysical Research: Solid Earth

(1978–2012), vol. 95, pp. 12653-12680, 1990.

[10] Abrams and Hook. (2002, 2008). ASTER User

Handbook,http://asterweb.jpl.nasa.gov/documents.asp (accessed 21 January 2008)

Available: http://asterweb.jpl.nasa.gov/documents.asp (accessed 21 January 2008)

[11] X. Zhang, M. Pazner, and N. Duke, "Lithologic and mineral information extraction for gold

exploration using ASTER data in the south Chocolate Mountains (California)," ISPRS

Journal of Photogrammetry and Remote Sensing, vol. 62, pp. 271-282, 2007.



123

[12] M. Abrams, S. Hook, and B. Ramachandran, "ASTER user handbook, version 2," Jet

propulsion laboratory, vol. 4800, p. 135, 2002.



[14] Application of remote sensing to hydrology; final technical report. Huntsville, Ala.: IBM

Federal Systems Division, reproduced by the National Technical Information Service,

1973.

[15] L. C. Rowan and J. C. Mars, "Lithologic mapping in the Mountain Pass, California area

using advanced spaceborne thermal emission and reflection radiometer (ASTER) data,"

Remote sensing of Environment, vol. 84, pp. 350-366, 2003.

[16] R. N. Clark, Spectroscopy of rocks and minerals and principles of spectroscopy. United

States (USA): John Wiley & Sons, New York, NY, United States (USA), 1999.

[17] R. N. Clark, G. A. Swayze, R. Wise, K. E. Livo, T. M. Hoefen, R. F. Kokaly, et al., USGS

digital spectral library splib06a: US Geological Survey Reston, VA, 2007.



[19] F. Van der Meer, "Spectral reflectance of carbonate mineral mixtures and bidirectional

reflectance theory: Quantitative analysis techniques for application in remote sensing,"

Remote Sensing Reviews, vol. 13, pp. 67-94, 1995/08/01 1995.

[20] L. C. Rowan, M. J. Kingston, and J. K. Crowley, "Spectral reflectance of carbonatites and

related alkalic igneous rocks; selected samples from four North American localities,"

Economic Geology, vol. 81, pp. 857-871, 1986.

[21] T. L. Bowers and L. C. Rowan, "Remote mineralogic and lithologic mapping of the ice

river alkaline complex, British Columbia, Canada, using AVIRIS data," Photogrammetric

Engineering and Remote Sensing, vol. 62, pp. 1379-1386, 1996.


Greenland, using HyMap imaging spectrometer data," Remote Sensing of Environment,

vol. 113, pp. 1208-1219, 2009.

[23] W. B. Ryan, S. M. Carbotte, J. O. Coplan, S. O'Hara, A. Melkonian, R. Arko, et al., "Global

Multi‐Resolution Topography synthesis," Geochemistry, Geophysics, Geosystems, vol. 10,

2009.


124

[24] M. Tarapoanca, P. Andriessen, K. Broto, L. Chérel, N. Ellouz-Zimmermann, J.-L. Faure,

et al., "Forward kinematic modelling of a regional transect in the Northern Emirates using

geological and apatite fission track age constraints on paleo-burial history," Arabian

Journal of Geosciences, vol. 3, pp. 395-411, 2010.

[25] "ASTER _07XT " NASA Land Processes Distributed Active Archive Center (LP

DAAC)vol. 2014, May 06, 2014, 3:46:30 PM 2014.

[26] F. A. Kruse, J. W. Boardman, and J. F. Huntington, "Comparison of airborne hyperspectral

data and EO-1 Hyperion for mineral mapping," Geoscience and Remote Sensing, IEEE

Transactions on, vol. 41, pp. 1388-1400, 2003.

[27] X. Chen, T. A. Warner, and D. J. Campagna, "Integrating visible, near‐infrared and short

wave infrared hyperspectral and multispectral thermal imagery for geologic mapping:

simulated data," International Journal of Remote Sensing, vol. 28, pp. 2415-2430, 2007.

[28] J. I. John R. Jensen, Perry Hardin, Ryan R. Jensen, Image Classification. The SAGE

Handbook of Remote Sensing. SAGE Publications Ltd. London: SAGE Publications Ltd.

[29] T. A. Warner, G. M. Foody, and M. D. Nellis, The SAGE handbook of remote sensing:

Sage Publications, 2009.

[30] J. F. Mustard and J. M. Sunshine, "Spectral analysis for earth science: investigations using

remote sensing data," Remote sensing for the earth sciences: Manual of remote sensing,

vol. 3, pp. 251-307, 1999.

[31] A. S. Mazer, M. Martin, M. Lee, and J. E. Solomon, "Image processing software for

imaging spectrometry data analysis," Remote Sensing of Environment, vol. 24, pp. 201-

210, 1988.

[32] F. Kruse, A. Lefkoff, and J. Dietz, "Expert system-based mineral mapping in northern

Death Valley, California/Nevada, using the airborne visible/infrared imaging spectrometer

(AVIRIS)," Remote Sensing of Environment, vol. 44, pp. 309-336, 1993.

[33] S. Kullback, Information theory and statistics: Courier Corporation, 1997.

[34] C.-C. Chang and C.-J. Lin, "LIBSVM: a library for support vector machines," ACM

Transactions on Intelligent Systems and Technology (TIST), vol. 2, p. 27, 2011.

[35] C.-I. Chang, "Spectral information divergence for hyperspectral image analysis," in

Geoscience and Remote Sensing Symposium, 1999. IGARSS'99 Proceedings. IEEE 1999

International, 1999, pp. 509-511.

125

[36] F. van der Meer, "The effectiveness of spectral similarity measures for the analysis of

hyperspectral imagery," International journal of applied earth observation and

geoinformation, vol. 8, pp. 3-17, 2006.

[37] F. Kruse, A. Lefkoff, J. Boardman, K. Heidebrecht, A. Shapiro, P. Barloon, et al., "The

spectral image processing system (SIPS)—interactive visualization and analysis of

imaging spectrometer data," Remote sensing of environment, vol. 44, pp. 145-163, 1993.

[38] O. A. De Carvalho and P. R. Meneses, "Spectral correlation mapper (SCM): an

improvement on the spectral angle mapper (SAM)," in Summaries of the 9th JPL Airborne

Earth Science Workshop, JPL Publication 00-18, 2000.

[39] G. P. Petropoulos, K. P. Vadrevu, G. Xanthopoulos, G. Karantounias, and M. Scholze, "A

comparison of spectral angle mapper and artificial neural network classifiers combined

with Landsat TM imagery analysis for obtaining burnt area mapping," Sensors, vol. 10, pp.

1967-1985, 2010.

[40] P. Mantero, G. Moser, and S. B. Serpico, "Partially supervised classification of remote

sensing images through SVM-based probability density estimation," Geoscience and

Remote Sensing, IEEE Transactions on, vol. 43, pp. 559-570, 2005.

[41] G. Mountrakis, J. Im, and C. Ogole, "Support vector machines in remote sensing: A

review," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 66, pp. 247-259,

2011.

[42] C. J. Burges, "A tutorial on support vector machines for pattern recognition," Data mining

and knowledge discovery, vol. 2, pp. 121-167, 1998.

[43] O. Chapelle, P. Haffner, and V. N. Vapnik, "Support vector machines for histogram-based

image classification," Neural Networks, IEEE Transactions on, vol. 10, pp. 1055-1064,

1999.

[44] G. Zhu and D. G. Blumberg, "Classification using ASTER data and SVM algorithms;: The

case study of Beer Sheva, Israel," Remote sensing of Environment, vol. 80, pp. 233-240,

2002.

[45] J. R. Jensen, Introductory digital image processing: a remote sensing perspective. Upper

Saddle River, N.J: Prentice Hall, 2005.

[46] G. M. Foody, "Status of land cover classification accuracy assessment," Remote Sensing

of Environment, vol. 80, pp. 185-201, 4// 2002.

[47] J. B. Campbell, Introduction to remote sensing: CRC Press, 2002.

126

[48] J. B. Campbell and R. H. Wynne, Introduction to remote sensing. New York: Guioford

Press, 2011.

[49] P. C. Smits, S. G. Dellepiane, and R. A. Schowengerdt, "Quality assessment of image

classification algorithms for land-cover mapping: A review and a proposal for a cost-based

approach," International Journal of Remote Sensing, vol. 20, pp. 1461-1486, 1999/01/01

1999.

[50] G. M. Foody, "Approaches for the production and evaluation of fuzzy land cover

classifications from remotely-sensed data," International Journal of Remote Sensing, vol.

17, pp. 1317-1340, 1996/05/01 1996.

[51] J. D. Wickham, R. V. O'neill, K. H. Riitters, T. G. Wade, and K. B. Jones, "Sensitivity of

selected landscape pattern metrics to land-cover misclassification and differences in land-

cover composition," Photogrammetric Engineering and Remote Sensing, vol. 63, pp. 397-

401, 1997.

[52] R. A. Monserud and R. Leemans, "Comparing global vegetation maps with the Kappa

statistic," Ecological modelling, vol. 62, pp. 275-293, 1992.

[53] J. R. Landis and G. G. Koch, "The measurement of observer agreement for categorical

data," biometrics, pp. 159-174, 1977.

[54] S. Stehman, "Comparing thematic maps based on map value," International Journal of

Remote Sensing, vol. 20, pp. 2347-2366, 1999.

[55] J. P. Rafferty, Minerals: Britannica Educational Pub., 2011.

[56] F. C. Hawthorne, A. Kato, H. J. Kisch, V. G. Krivovichev, K. Linthout, J. LAIRD, et al.,

"Nomenclature of amphiboles: report of the subcommittee on amphiboles of the

International Mineralogical Association, Commission on New Minerals and Mineral

Names," Can Mineral, vol. 35, p. 219246Lindsley, 1997.

[57] F. A. Kruse and S. L. Perry, "Mineral mapping using simulated Worldview-3 short-wave-

infrared imagery," Remote Sensing, vol. 5, pp. 2688-2703, 2013.

[58] E. Bedini, F. Van Der Meer, and F. Van Ruitenbeek, "Use of HyMap imaging spectrometer

data to map mineralogy in the Rodalquilar caldera, southeast Spain," International Journal

of Remote Sensing, vol. 30, pp. 327-348, 2009.

127

4. CHAPTER 4

DERIVING THE GEOCHEMICAL CHARACTERISTIC OF

CARBONATITES USING VISIBLE NEAR INFRARED AND SHORT

WAVE INFRARED SPECTROSCOPY

Abstract

Rare Earth Elements (REE) have become an increasingly important natural resource

considering their central role in the development of new green technologies. Locating REE

deposits that are economically feasible for exploitation is a challenge. In this investigation,

spectroscopic methods are utilized for the identification of carbonatite, an important host of REEs.

The visible and near infrared (VNIR) and short wave infrared (SWIR) wavelength band of two

common varieties of carbonatite (ferro- and calcio-carbonatite) were examined to correlate the

spectral absorbances with the carbonatite’s geochemical properties. Through our investigation, we

found that carbonatites have three key geochemical components which yield distinctive and

correlatable spectral signatures (calcite, iron, and, neodymium). Calcite is a major mineral within

carbonatite, and displays vibrational absorption between 2.312–2.345 μm for both fresh and

weathered surfaces of all carbonatite samples we tested. The absorption depth is variable and

generally depends on the calcite concentration within a sample. The most pronounced calcite

absorption was recorded from a fresh surface of a calcio-carbonatite sample at 2.33 μm. REE (Nd)

absorption, is evident in both ferro- and calcio-carbonatites between 0.7 to 0.774 µm. However,

128

the most pronounced absorption (0.742 µm) is observed for fresh surface samples of ferro-

carbonatite; calcio-carbonatites exhibit less absorption on both weathered and fresh surfaces. Fe-

oxides exhibit absorption within the VNIR region between 0.866-1.69 μm, with the most

pronounced iron oxide absorption recorded from the fresh surface of a ferro-carbonatite at 1.167

μm. In this research, geochemical analysis was utilized to enhance the interpretations of spectral

data by examinaing the correlation between spectral absorption and elemental abunadance.

Spectral absorption depths for calcium, neodymium, and iron of fresh and weathered carbonatite

surfaces were plotted against geochemically determined concentrations of these elements derived

by XRF and LA-ICPMS analysis of the same sample billets. The results show a positive relation

(R2 0.8 to ≥ 0.9) between absorption depths and the carbonatite's geochemcial composition. We

examine the relationship of carbonatite Nd absorption spectra with the available published data

about Nd absorbtion in world carbonatites and find significantly more scatter, with values of R2 of

~0.6. While there is some degree of positive correlation between absorption depths and Nd

concentration within carbonatites, the scatter reflects complexities in making such spectral-

geochemical correlations. In particular, the difference between a carbonatite fresh interior and

weathered exterior spectra may result in a less robust correlation. Overall, our results suggest that

it is possible to derive the geochemistry of carbonatite key indicators (calcite, neodymium, and

iron abundances) using a spectroradiometer instrument and methods if care is taken to consider

both fresh and weathered carbonatite surfaces, as well as the heterogeniety of the sample.

129

1. Introduction

The growth of new green technologies has been accompanied by an increasing industrial

demand for rare earth elements (REEs) – critical components in high-efficiency engines. Global

REE production comes from just a few locations, as a consequence, locating new exploitable

sources of REEs has become a critical economic and security priority [1]. Most exploitable

deposits of REEs are hard to find and usually concentrate within REE-bearing minerals in

carbonatites [2]. Remote sensing provides a mechanism by which new deposits of carbonatite may

be identified, though challenges remain in constraining the spectral characteristics of these rocks.

Mapping carbonatite without reference spectra is often complicated by the different

minerals/elements present, and the limited outcrop footprint of carbonatites, which are often

embedded within large-scale geological units [3]. Examination of previous studies that sought to

identify carbonatite by spectroscopic approaches reveals key outstanding questions that remain

unresolved: i) What are the suitable key minerals/elements that can be used to unambigiously

indicate carbobnatite? ii) What is the difference between a carbonatite fresh interior and weathered

exterior spectra? iii) Does the grain size and carbonatite homogeneity affect the spectral behavior?

iv) How do the spectral features correlate with carbonatite REE concentrations? The establishment

of carbonatite reference spectra, and an examination of its detailed characteristics can improve on

the existing difficulties in mapping of carbonatite. To address these issues, we carefully examined

selected samples (for both fresh interiors and weathered exteriors) of REE-enriched carbonatite

from the Al-Uyaynah fields (UAE), using both spectroscopy to assess the spectral characteristics

of the rocks and inductively coupled plasma mass spectrometry and X-ray fluorescence (XRF)

spectrometer to determine the geochemical composition of the rocks.

130

The main objectives of research are: (i) to develop an understanding of the spectral

charachterisitcs of the carbonatites spectroscopy by examining their spectral reflectance within

visible, near infrared (VNIR) and short wave infrared (SWIR) bands utilizing an analytical spectral

device (ASD), (ii) acquiring a more comprehensive understanding of the geochemical properties

of the Al-Uyaynah carbonatites by using geochemical analysis methods (XRF and LA-ICP-MS)

on the same samples, and (iii) testing the linearity of the relationship between the absorption

features of carbonatite spectra and their concentration in carbonatites by chemically analyzing the

same material as was charachterized spectrally. This approach miminizes the potential for

mismatches caused by inhomogenity within a carbonatite surface (e.g. presence of absence of large

REE-bearing crystals). This experiment presents the laboratory reflectance spectra of ferro- and

calcio-carbonatites that were collected from the well-exposed Al-Uyaynah area-fields located in

the UAE. The new carbonatite spectra may contribute novel Nd-rich spectra increasing our

understanding of carbonatite spectroscopy, and also expanding the spectral data to the

spectroscopy community. These data provide the first comprehensive analysis of carbonatite

spectra from both fresh and weathered surfaces of carbonatities and has broad application to the

identification of carbonatites by remote sensing methods.

2. Background

Spectroscopy is the discipline of studying the interaction between matter and light [4].

Usually, the interaction with matter generates a range of observable optical/physical phenomena

such as absorption, transmission, and reflection giving information about minerals/elements of the

parent rocks [5]. A number of authors [6-8] have carried out research on rock-forming minerals by

131

employing absorption properties of reflectance spectra in the visible and near infrared (VNIR) and

shortwave infrared (SWIR) bands (Figure 4.1). The absorption in the spectral wavelength is caused

by two general processes: 1) electronic processes and 2) vibrational processes [9]. The

mineral/element absorption features in visible and near infrared (VNIR) bands are mostly specified

by the electronic transitions or processes like charge transfer effect, and crystal field effect [10-

12]. The spectra absorption features result from the change in state of energy that follows the

absorption of a photon. Meanwhile the atoms and ions have distinct states of energy within the

crystal lattice that vibrate in response to incident radiation creating absorption in SWIR bands[5,

13]. Furthermore, the spectral absorption features of minerals are caused by the partially filled

electron shells (crystal field effect) of transition elements like Fe, Co, and Cr, which is the main

electronic process that controls the absorption properties in minerals [5, 8, 13]. Vibrational

processes cause the absorption characteristics of mineral spectra at specific locations within the

SWIR and TIR bands [5, 13, 14]. Electronic transitions appears to govern mineral spectral

properties at VNIR wavelength areas more than vibrational processes [15]. The vibrational

processes happen mainly because of vibrational movement in the mineral crystal lattices, like,

minor displacements of the atoms regarding their resting states [16]. Three modes are present with

vibrational processes and include overtone, combination, and fundamental vibrations [5, 13].

Generally, vibrational processes are a result of the occurrence of hydroxyl ion (OH-), carbonate

(CO32-), water (H2O), and other relevant minerals [18].

In analyzing the composition of minerals/elements on the basis of absorption

characteristics, the VNIR and SWIR bands of the reflectance spectrum are quite important [19-

21]. Carbonatites, which are enriched with about 50% or more of carbonate minerals, exhibit a

132

Figure 4.1. The range of the electromagnetic spectrum, showing most crucial ranges for minerals

sensing spectroscopy applications. The visible/ near infrared (VNIR), and short-wave infrared (SWIR),

modified from [17].

133

particularly strong absorption characteristic of the CO32- ion detected readily within the range of

2.3-2.35 µm of SWIR bands [5, 22]. The VNIR bands’ data typically are an invaluable source of

information regarding iron (Fe2+ and Fe3+) [5, 16, 19]. Inside the VNIR bands of wavelengths

between 0.8-0.9 µm the absorption features of iron-rich carbonatites are pronounced [23-25].

Some carbonatites contain significant amounts of rare earth elements (REE), like cerium (Ce),

europium (Eu), neodymium (Nd), and lanthanum (La) [2]. The REE spectrum is largely related to

electronic transitions, which are clearly shown by the spectra absorption properties in the VNIR

wavelength in the area between 0.5-0.8 um [20, 26].

The spectral absorption properties of carbonatites have been used in previous studies to

create maps of carbonatites, which are measurable from aircraft reflectance spectroscopy. For

example, AVIRIS (Airborne Visible / Infrared Imaging Spectrometer) images were used to

identify the Ice River Carbonatite complex and showed minor absorption features located at 0.58,

0.74, 0.80, and 0.87 μm, which is characteristic of Nd3+ [27]. Bedini and Tukiainene employed

spectroscopic spectra to distinguish between carbonatite, fenite, and hematised surface rock, with

particular attention to the Nd3+ reflectance characteristics of Sarfartoq carbonatite in Greenland

[28]. Recent contributions [29] discussed the possibility of spectroscopic approaches

(hyperspectral remote sensing), within VNIR-SWIR wavelengths to distinguish neodymium in

carbonatite and alkaline igneous rocks that accommodated REEs. These authors analyzed

collections of carbonatite which were obtained from previous research and museum collections of

carbonatites with a wide geographic distribution. This study reported that Nd absorption features

in carbonatite is the most pronounced among the REE absorption characteristics in the VNIR-

SWIR spectrum.

134

3. Study area location and geology

The Al-Uyaynah fields are located approximately 18 km to the southwest of Dibba city, in

Fujairah state of the UAE. The Al-Uyaynah fields contain extrusive carbonatites, which lie within

the metamorphic mélange unit to the west side of the Semail ophiolite complex. These carbonatites

have been previously studied and interpreted as sub-aerially erupted volcanics within oceanic

metasediments as an accreted volcanic island predating obduction of the Semail Ophiolite

Complex [3]. The Al-Uyaynah carbonatites are comprised of ~10-50 m thick layers with abundant

lapilli. The presence of the lapilli requires a sub-aerial origin. The carbonatites are described as

being composed of two typical magmas [3]. The Type 1 carbonatite contains calcite as the main

mineral phase and is generally comparable to calcio-carbonatites. These carbonatites also contain

abundant apatite and opaque minerals such as magnetite. The typical composition

of these lapilli is 40% calcite, 20-30% apatite and opaque minerals [3]. The Type 2 carbonatite

contains higher modal abundances of silicates in comparison Type 1 (calcio-carbonatites). The

representative minerals are calcic, amphibole, biotite, and abundant of opaque minerals such as

magnetite and is geochemically equivalent to ferro-carbonatites. Allanite in ferro-carbonatites

usually occurs as crystal accumulations. These accumulations concentrate REEs and result in trace

elements heterogeneously distributed within the rock [3].

135

Figure 4.2. A field view of Al-Uyaynah ferro-carbonatites, (a) shows the ferro-carbonatites outcrops

of the northern end of the study area, (b, and c) show a close view of ferro-carbonatites. (d) hand

sample from ferro-carbonatite area.

136

Figure 4.3. A field pictures of Al-Uyaynah calcio-carbonatites (a) shows the calcio-carbonatites

outcrops of the northern part of the study area, (b, and c) show a close view of clacio-carbonatites. (d)

a hand sample calcio-carbonatite.

137

4. Methods

4.1. Sample preparation

Samples of carbonatites were collected from the Al-Uyaynah field during a field expedition

in December 2010 (Fig. 4.2 and 4.3). The samples were prepared in the research facilities of the

Department of Geological Sciences at Michigan State University. Each carbonatite sample was cut

into rectangular cubes using a diamond slab saw, with care taken to preserve the weathered

surfaces, and to exposure fresh interior surfaces. The interior surfaces of billets were polished

using a rotary sanding plate. Billets were washed using an ultrasonic cleaner filled with deionized

water to eliminate unwanted materials like clay, dust, and polishing grit. The cleaned samples

were then placed on a hot plate at about 60°C to dry them off.

4.2. Spectral Analysis using ASD

The prepared sample billets were analyzed using an ASD FieldSpec-3 Hi-Res

Spectroradiometer located at the Department of Environmental Sciences at the main campus of the

Univesity of Toledo. The Illumination source was provided by a 1000W 3200K Quartz Halogen

lamp with a 10-degree foreoptic. For each sample, reflectance spectra were obtained for both

unweathered and weathered surfaces over the wavelength range from 0.35 to 2.5 μm. The spectral

resolution of ASD is 3 nm for VNIR band, and 8 nm for SWIR band. All spectra were normalized

to a Spectrlon® lab standard.

138

To normalize the carbonatite laboratory spectra, we utilized the continuum removal method

[30]. Continuum removal is a method of reflectance spectra normalization that reduces the spectral

albedo background or continuum, and is computed by dividing every spectrum by the

corresponding continuum line [30]. This technique improves differences in the absorption shape

of individual mineral properties [31], and permits a comparison of individual absorption features

from other spectral mineral databases. We utilized the continuum removal tool, (PRISM version

1.0) [32], from USGS and which is embedded in the ENVI software system [33]. The tool was

utilized to normalize carbonatite spectra and to distinguish the accurate positions of the absorption

center, and to measure the absorption depth of formed by each mineral/element. The wavelength

of the laboratory reflectance spectra is shown in Figure 4.4.

4.3. Geochemical analysis

The same samples that were used in spectroscopy analysis were then re-polished gently

and cleaned again after the spectra measurement was completed. The billets were crushed in a steel

jaw crusher and hand powdered using agate mortar and placed in glass container. Samples were

placed in a vacuum oven at 110oC for a minimum of 24 hours. These powdered samples were

blended with dehydrated lithium tetraborate (Li2B4O7) using methods described in [34] to make a

glass disk. The mixture ratio (3:9 or 1:9) was based on the powdered samples' volume. To ensure

full oxidation during sample fusion, an oxidizing agent, ammonium nitrate (NH4NO3; 0.5g) was

added. The final mixtures were subsequently heated and melted in platinum crucibles and gently

agitated in an orbital mixer at 1000oC until homogeneous melts were achieved. The melts were

then poured into platinum moulds and transferred to a hot plate to anneal, then to a ceramic plate

139

Figure 4.4. Example of the field spectra of ferro-carbonatite. The original reflectance graph in red,

and the reflectance after applying the continuum removal method in green. Absorption depth, area,

and center are labelled.

140

to cool to room temperature and form glass disks. These glass disks made from carbonatite

powders were subsequently analyzed for major oxide concentrations in weight percent of whole

rock. This was accomplished by utilizing a Bruker S4 Pioneer wavelength dispersive X-ray

fluorescence (XRF) spectrometer at Michigan State University, and analyzed using the silicate

LDF method, following the protocol of [34].

The trace elements, including rare-earth element concentrations in parts per million (or

ppm), were derived using a laser ablation inductively coupled plasma mass spectrometry (LA-ICP-

MS) at Michigan State University's geochemistry lab on the same samples. The laser ablation

analysis was performed based on the methods described in [35]. The laser ablation was conducted

for 180 seconds as surface scans at 10 m /s with a diameter about 110 m, operating a Photon-

Machines Analyte (G2 193 nm) excimer laser set at 4.1 mJ/cm-2 fluence and 10 Hz repetition rate.

The carbonatite glass samples were placed within a 1515 cm HelEx sample chamber, which is

capable of operating about 30 mounted samples and standards in one processing. The chamber

containing glass samples was evacuated and a helium (0.75 L/min high-purity at 1 atm) carrier gas

flux was used to carry the ablated material to the ICPMS. Shortly before analysis, samples were

pre-ablated at 100 m/s with a 150 m beam to avoid or eliminate possible contamination of

sample surfaces. To decrease molecular interferences on light masses and enhance the detection

limits, the ICPMS (Thermo Scientific ICAP Q quadrupole) was set to kinetic energy dispersion

(KED). The LA-ICP-MS data were analysed using Qtegra software supplied by Thermo to extract

the REE concentrations. The results of XRF and LA-ICP-MS analysis results are illustrating in

tables 4.1 and 4.2.

141

5. Results

5.1. Carbonatite Spectra

5.1.1. Calcite related absorption

Results show, broadly, that the spectra are dominated by calcite, iron, and REEs. Carbonate

minerals have systematic vibrational absorption peculiarities in the SWIR band, that can be utilized

for distinguishing the influence of carbonate minerals on bulk rock spectra bands [5, 22]. These

vibrational absorption bands are evident in carbonate spectra because internal vibrations of the C-

O bond within the SWIR band occurs in the area between 2.30–2.35 μm [22, 23].

Within our study, calcite is the dominant phase in both calcio- and ferro-carbonatites,

however within calcio-carbonatite, it is more prevalent comprising around 40% of its lapilli

compositions [3]. The wavelength area that correlates to calcite absorption, was measured based

on the normalized laboratory spectra (continuum removal) results of both ferro- and calcio-

carbonatite samples. The absorption center was observed between 2.312–2.345 μm for fresh

interior and weathered exterior of all carbonatite samples (Figure 4.5 and 4.6). The most

pronounced or deepest absorptions were recorded at 2.33 μm of the calcio-carbonatite fresh interior

surface. In contrast, a less significant absorption was found in this area for the interior surfaces of

142

Figure 4.5. Shows calcite absorption bands in USGS spectra library (left) [36] , and calcite absorption

area which recorded in calcio-carbonatite (right). The absorption was observed in the area between

2.312–2.345 μm, and the deepest absorption point was recorded at 2.34 μm of fresh calcio-carbonatite.

143

Figure 4.6. Illustrates calcite absorption bands in ferro-carbonatites. The absorption was between 2.312–

2.345 μm of the spectra wavelength, and the deepest absorption point was at 2.34 μm of much of ferro-

carbonatites.

144

ferro-carbonatite samples due to the relatively lower calcite abundance (Figure 4.6). The

weathered exteriors of calcio- and ferro-carbonatite display a somewhat less pronounced

absorption in comparison to the inner surfaces. This observation is consistent with our observations

of less carbonate on these weathered exterior surfaces. The correlation between the calcite

concentration and wavelength absorption depths were plotted (absorption depth vs %) and show a

positive linear correlation with an R2 value of 0.8 (Figure 4.14).

5.1.2. Iron related absorption

Iron oxides have diagnostic electric processes or charge transfer absorption features in the

VNIR band [5, 16, 19]. Many scholars have recorded the absorption features of iron oxides, (e.g.

[21]), within the visible near infrared ranges (VNIR). They report the pure iron oxide absorption

feature in the area between 0.9-1.0 μm. Fe2+ displays absorption near 1.0 μm, while minerals that

exhibit Fe3+ absorption characteristics (e.g., hematite), have a diagnostic absorption around 0.9 μm

[5].

In our sample suite, ferrous and ferric ions (Fe2+, Fe3+) occur within iron oxide minerals in

ferro-carbonatites and also in the weathered exterior of calcio-carbonatites. This charachteristic is

quite noticeable in the field, and manifests as a coating of iron oxdies showing a rusted appearance

on the weathered exterior of the carbonatites. In the ferro-carbonatites iron-oxide minerals

(magnetite and hematite) are important phases, whereas iron-oxide minerals are just a subordinate

phase in calcio-carbonatites [3].

145

The normalized spectra (continuum removal normalized) of the ferro and calcio-

carbonatite samples both exhibit absorption in the VNIR region, but values for the absorption

center vary due to spectral impurity recorded between 0.885-1.037μm. Iron-rich minerals

(magnetite and hematite) are abundant in the ferro-carbonatite fresh interior more so than the

clacio-carbonatite (though weathered surfaces of both rock types exhibited such minerals). The

deepest absorptions were observed at 0.885 μm and around 0.907 μm for ferro-carbonatite (Figure

4.7, and 4.17). Iron oxide concentrations are greater in the weathered/fresh surface of ferro-

carbonatites as compared to calcio-carbonatites, and these absorption features can be applied in

differentiating between the spectra of the two carbonatites. However, carbonatite spectra are

unusual and spectral impurity is evident. This causes slight shifting of the absorption center in

comparison to that reported for pure iron oxides in the literature. Overall the Fe absorption depth

in carbonatites was well correlated with concentration (R2 > 0.8) (Figure 4.18).

5.1.3. Rare earth element (REE) related absorption

The rare earth element (REE) absorption bands are defined via electronic field transition

absorption features [26]. The REEs spectral absorption is commonly measurable within the visible

region [21], which is clearly shown by absorption features in the VNIR wavelength ranges between

0.5- 0.87 um [20, 26]. Neodymium (Nd) absorption features are most notable among the REE,

therefore, Nd is one of the key elements for the entire REE group [29].

In the Al-Uyaynah carbonatites, ferro-carbonatites have higher concentrations of REE-

bearing minerals in their lapilli, while calcio-carbonatite have considerably lower concentrations.

146

Figure 4.7. Shows iron oxides from USGS spectra library (left) [36] , and the iron oxide absorption area that is observed in ferro-carbonatite

(middle). The iron absorption was recorded in the area between 0.8-1.26μm. The deepest absorption point was recorded around 0.885 μm of

fresh ferro-carbonatite and 1.037 μm for the weathered part. The calcio-carbonatite shows shallow absorption in both fresh and weathered

surfaces, however, the iron oxide absorption was more pronounced in weathered outcrop.

147

Figure 4.8. Shows neodymium (Nd) from USGS spectra library (left) [36], and the Nd absorption area

that was observed in ferro-carbonatite (right). The Nd absorption spanned the area from 0.7-0.78μm. The

deepest absorption was recorded around 0.74 μm.

148

Figure 4.9. Neodymium (Nd) absorption area that was observed in ferro-silico-carbonatite weathered

surface (left) and fresh interior (right). The Nd absorption depth is higher (at 0.705 µm) in fresh surfaces

in comparison to weathered surfaces.

149

Figure 4.10. Nd absorption in calcio-carbonatite was verified. The deepest absorption point was recorded

around 0.752μm of the weathered surface of the calcio-carbonatite spectrum. The un-weathered part

shows a shallower absorption at 0.751 μm.

150

The normalized spectra of ferro- and calcio-carbonatite samples indicate REE-related absorption

features in the VNIR wavelength range between 0.7 to 0.774 µm (Illustrated in Figure 4.8 and

4.10). The Nd absorption depth is variable, as the impact of iron interference is known to decrease

the Nd absorption features [37] in some specimens. The deepest point of absorption is for fresh

interior surfaces of ferro-carbonatite, and is observed at 0.705 µm (Figure 4.9). The weathered

exterior surface of ferro-carbonatite exhibits a spectrum with a similar Nd absorption center

between 0.705 - 0.742 µm (Figure 4.15). However, in calcio-carbonatites, the normalized ASD

spectra shows a Nd absorption center at 0.751 µm for the fresh interior surface, and at 0.752 for

the exterior weathered surface. The calcio-carbonatites’ Nd absorptions are less pronounced than

those recorded in the ferro-carbonatite spectra due to of the lower Nd concentration in the calcio-

carbonatite. The coefficient of determination of the Nd concentration vs its absorption depth

exhibits a positive correlation (R2> 0.9 for Nd), (Figure .16).

5.1.4. Major element analysis by XRF

For the purpose of linking the spectral data to the viability of discriminating carbonatites

through spectral analyses, the same representative set of samples that were collected from the Al-

Uyaynah carbonatite were chemically analyzed. The major oxides were determined by using the

XRF Silicates LDF method noted above.

Woolley and others conducted geochemical research on the carbonatites of the Al-Uyaynah

area during 1991. They divided the Al-Uyaynah carbonatites based on petrologic features into two

distinct groups: i) Type 1, which is similar in composition to average calcio-carbonatite; ii) Type

151

2 is a silico-carbonatite which is chemically similar to average ferro-carbonatites (here we use the

term ferro-carbonatite). The geochemical data determined from this study are shown in Table 4.1.

Included in this table are data of average Type 1 and average Type 2 carbonatites from Woolley

(1991) and worldwide average calcio- and ferro-carbonatite compositions.

The results obtained from XRF analysis of our samples as major oxides distribution in wt.

% were compared with the compositions of published samples presented by Wooley, which show

a very analogous composition to the average of carbonatite. The average chemical data of our

samples also show clearly two groups of carbonatites, calcio and ferro-carbonatite, that falls within

the ranges that are stated in literature.

5.2. Geochemical impact of weathering on major element composition

The variation between the fresh and weathered surface of both varieties of carbonatite were

examined. The CaO concentrations are about 42 wt. % for weathered to 45 wt. % for fresh part of

calcio-carbonatite, suggesting some CaO loss. The CaO concentrations in ferro-carbonatite

samples are universally lower than calcio-carbonatite, however these too exhibit CaO less in

weathered surfaces (26-29 wt.%) in comparison to fresh interiors (29-29.9 wt.%) (Figure 4.11).

The weathered rind of all samples are dominated by Fe-oxides. The Fe2O3 concentration is about

21wt.% in the ferro-carbonatite weathered surface, while only 17 wt.% in the fresh interior.

Similarly, the calcio-carbonatite weathered exterior also exhibits higher concentrations of Fe2O3

(13%) in comparison to its fresh interior (10%). SiO2 concentrations do not show as clear a

variation, in weathered ferro-carbonatite SiO2 ranges from 13-21wt.%, while its between 15-24

152

Ferro/silico-carbonatite

Ferro-carbonatite

Calcio-carbonatite

Figure 4.11. Illustrates The CaO, Fe2O3 and SiO2 concentration in the weathered exterior and fresh interior

from selected field samples of Al-Uyaynah carbonatite. The CaO is higher in calcio-carbonatite, while the iron

oxides are greater in ferro-carbonatites. The weathered part of carbonatites contains more iron oxides than the

fresh part of the rock. SiO2 has higher concentrations in ferro-carbonatite samples versus calcio-carbonatites.

1 1_1 2 2_1 4 4_1 27 27_1 28 28_1 5 5_1

SiO2 (%) 24.34 21.89 23.52 35.5 15.07 16.24 14.91 13.53 11.45 15.77 3.02 3.6

FW

F

W

FW

FW

F

W

F W

0

5

10

15

20

25

30

35

40

SiO

2 w

%

1 1_1 2 2_1 4 4_1 27 27_1 28 28_1 5 5_1

Fe2O3 (%) 8.91 9.23 9.05 12.98 17 16.1817.1821.83 18.2 18.5910.2613.53

F W F

W

FW

F

W

F W

F

W

0

5

10

15

20

25

Fe 2

O3

W%

1 1_1 2 2_1 4 4_1 2727_

128

28_

15 5_1

CaO (%) 26.2 29.7 27.8 14.5 29.5 29.3 29.9 26.7 31.4 27 45.3 42.2

F

WF

W

F W F

W

F

W

F

W

0

5

10

15

20

25

30

35

40

45

50

CaO

W%

153

wt.% in fresh ferro-carbonatite samples. The SiO2 concentration in the calcio-carbonatite

weathered surface is 3.6 wt.%, and is 3 wt.% in the un-weathered fresh interior of the calcio-

carbonatite. MgO, Na2O, and K2O concentrations are higher in weathered ferro- and calcio-

carbonatites versus the fresh interior.

5.3. CIPW normative mineral composition

CIPW normative mineral compositions were calculated for Al-Uyaynah carbonatite samples from

the whole rock geochemical data. Calculations were performed using a spreadsheet macro

prepared by Kurt Hollocher after Johannsen, 1931. The mineral assemblages are shown in Table

4.3. All samples possess the same basic calculated mineral assemblage but have variation of weight

percentage of the different phases within the assemblage. Calcio-carbonatite (samples 5 and 5-1)

have the highest calcite at 66% and 55%, respectively. The amount of apatite is 15.39% for fresh

and 18.52% for weathered portion. Fe-oxides are second lowest. Ferro-carbonatite samples

(samples 1, 2,4,27 and 28) have lowest calcite content. Correspondingly they have higher content

of silicate minerals and also have the highest amount of apatite (21-27%). These samples also have

the highest amount of Fe-oxides. In all samples, calcite is slightly lower in concentration for the

weathered portion of the rock. The Fe-oxides exhibit the same trend of being slightly increased in

the weathered portion over the fresh rock.

154

5.4. Trace element analysis by LA-ICPMS

In 1984, Cullers and others undertook a detailed study of REE concentrations in carbonatites, and

pointed out that carbonatites have the highest concentration of REEs of any igneous rocks [38, 39].

Furthermore, there was predictable variability between carbonatite varieties, whereby the

concentration of REEs increase progressively from calcio-carbonatite, to magnesio-carbonatite, to

ferro-carbonatite [39]. Our results are broadly consistent with these global observations - ferro-

carbonatites from the Al-Uyaynah carbonatite are significantly more enriched in REE in

comparison to calcio-carbonatites. Ferro-carbonatite samples (1, 1-1, 4, 4-1, 27, 27-1) show overall

greater enrichment in REEs to calcio-carbonatite samples (5, 5-1) and also possesses a greater

LREE to HREE enrichment.

The REE results from this study are compared with existing REE data from the Al-

Uyaynah carbonatites [39] and the world average values of calcio- and ferro-carbonatites (Fig

4.12). Our data show good agreement with the two types of carbonatites noted in prior studies

[39]. Furthermore, our ferro-carbonatite samples (greater LREE to HREE enrichment and overall

enrichment) correspond well with the global average value for ferro-carbonatite. Similarly, calcio-

carbonatites from our study agree with global average values for calcio-carbonatite. The ICP-MS

analyses of Al-Uyaynah carbonatite interior and weathered surfaces generally display enrichment

in LREEs concentrations in the weathered portion over fresh interior (Table 4.2).

155

the major oxides distribution of selected Al-Uyaynah carbonatite samples evaluated using XRF. All carbonatite samples contain calcium concentrations 26-45 Wt.

%. The lower concentrations of carbonate were recorded in ferro-carbonatites fresh and weathered samples and those ferro-carbonatites exhibit a higher abundance

of iron and silica in comparison to calcio-carbonatites.

Sample

Name SiO2 TiO2 Al2O3 Fe2O3 MnO MgO CaO Na2O K2O P2O5 LOI

Total

(%)

1 Ferro-

carbona

tite

Rich in

silicates

24.34 0.25 1.67 8.91 0.23 5.09 26.15 1.81 0.09 1.41 28.42 98.37

1_1 21.89 0.31 1.77 9.23 0.48 4.46 29.65 1.28 0.2 4.17 25.31 98.75

2 23.52 0.19 1.6 9.05 0.54 4.89 27.79 1.67 0.07 1.31 27.96 98.59

2-1 35.5 0.31 2.46 12.98 0.3 7.37 14.48 2.5 0.12 2.16 20.16 98.34

4 Ferro-

carbona

tite 15.07 1.29 2.21 17 0.43 8.12 29.5 0.02 0.01 9.65 16.1 99.4

4-1 16.24 1.26 2.29 16.18 0.46 8.5 29.33 0.03 0.03 9.36 15.79 99.47

5

Calcio-

carbona

tite 3.02 0.77 1.1 10.26 0.6 1.76 45.27 0 0 6.73 29.89 99.4

5-1

3.6 0.96 1.34 13.53 0.53 2.08 42.19 0 0.01 8.19 26.83 99.26

27 Ferro-

carbona

tite 14.91 1.31 2.18 17.18 0.43 8.05 29.87 0.02 0.01 9.44 15.99 99.39

27-1 13.53 1.65 3.09 21.83 0.52 7.87 26.66 0.04 0.02 12.26 11.88 99.35

28 11.45 1.4 2.39 18.2 0.41 6.71 31.38 0.02 0.01 10.32 17.06 99.35

28-1 15.77 1.42 2.65 18.59 0.46 8.62 27.04 0.03 0.02 10.56 14.23 99.39

Table 4.1. Illustrates the major oxides distribution of selected Al-Uyaynah carbonatite samples evaluated using XRF. All carbonatite samples contain calcium

concentrations 26-45 Wt. %. The lower concentrations of carbonate were recorded in ferro-carbonatites fresh and weathered samples and those ferro-carbonatites

exhibit a higher abundance of iron and silica in comparison to calcio-carbonatites.

156

Shows the distributions of REEs of selected carbonatite field samples from Al-Uyaynah field. The REE (ppm) were analyzed by (LA-ICP-MS).

Table 4.2. Shows the distributions of REEs of selected carbonatite field samples from Al-Uyaynah field. The REE (ppm) were analyzed by (LA-ICP-MS).

157

Input analysis Units 28 28_1 27 27_1 5 5_1 4 4_2 2 2_1 1 1_1

SiO2 % 11.45 15.77 14.91 13.53 3.02 3.6 15.07 16.14 23.52 35.5 24.34 21.89

TiO2 % 1.4 1.42 1.31 1.65 0.77 0.96 1.29 1.46 0.19 0.31 0.25 0.31

Al2O3 % 2.39 2.65 2.18 3.09 1.1 1.34 2.21 2.57 1.6 2.46 1.67 1.77

Fe2O3 % 18.2 18.59 17.18 21.83 10.26 13.53 17 18.92 9.05 12.98 8.91 9.23

FeO %

MnO % 0.41 0.46 0.43 0.52 0.6 0.53 0.43 0.48 0.54 0.3 0.23 0.48

MgO % 6.71 8.62 8.05 7.87 1.76 2.08 8.12 8.7 4.89 7.37 5.09 4.46

CaO % 31.38 27.04 29.87 26.66 45.27 42.19 29.5 27.12 27.79 14.48 26.15 29.65

Na2O % 0.02 0.03 0.02 0.04 0 0 0.02 0.04 1.67 2.5 1.81 1.28

K2O % 0.01 0.02 0.01 0.02 0 0.01 0.01 0.02 0.07 0.12 0.09 0.2

P2O5 % 10.32 10.56 9.44 12.26 6.73 8.19 9.65 10.73 1.31 2.16 1.41 4.17

CO2 % 17.06 14.23 15.99 11.88 29.89 26.83 16.1 13.23 27.96 20.16 28.42 25.31

SO3 %

S %

F %

Cl %

Sr ppm 3042 2283 2803 2304 3288 3970 2566 2177 2819 1501 3950 2913

Ba ppm 136 279 211 269 598 671 238 321 278 122 96 636

Ni ppm 33 39 33 42 13 16 33 38 381 459 330 241

Cr ppm 11 21 12 26 37 16 20 0 277 285 196 152

Zr ppm 1904 1886 1771 2194 1248 1604 1758 1953 445 513 418 595

Norms

Quartz 1.0807 2.3995 2.5131 1.3012 0.0739 0.0717 2.5398 2.6508 14.579 21.203 14.715

5

13.263

6

Plagioclase 0 0 0 0 0 0 0 0 0 0 0 0

Orthoclase 0.11 0.2201 0.1407 0.2172 0.2392 0.3231 0.1501 0.2391 0.4847 0.6714

8

0.5144

3

1.3418

9

Table 4.3. Shows the CIPW normative mineral compositions were calculated for Al-Uyaynah carbonatite samples from the whole rock XRF geochemical

data.

158



Nepheline 0 0 0 0 0 0 0 0 0 0 0 0

Leucite 0 0 0 0 0 0 0 0 0 0 0 0

Kalsilite 0 0 0 0 0 0 0 0 0 0 0 0

Corundum 2.2815 2.4815 2.0958 2.9134 1.0416 1.2487 2.1046 2.4314 1.3847 2.0544

7

1.4109

1

1.4039

5

Diopside 0 0 0 0 0 0 0 0 0 0 0 0

Hypersthene 16.095 20.432 19.513 18.734 4.3256 5.0568 19.512 20.871 11.217 16.248

6

11.426

4

10.354

Wollastonite 0 0 0 0 0 0 0 0 0 0 0 0

Olivine 0 0 0 0 0 0 0 0 0 0 0 0

Larnite 0 0 0 0 0 0 0 0 0 0 0 0

Acmite 0 0 0 0 0 0 0 0 0 0 0 0

K2SiO3 0 0 0 0 0 0 0 0 0 0 0 0

Na2SiO3 0 0 0 0 0 0 0 0 0 0 0 0

Rutile 0.9 0.8548 0.8001 1.0136 0.094 0.3534 0.7744 0.8805 0 0 0 0

Ilmenite 0.8513 0.9431 0.9018 1.0698 1.2644 1.1084 0.8928 0.9984 0.3323 0.5211

4

0.4279

3

0.5487

6

Magnetite 0 0 0 0 0 0 0 0 1.1977 0.1756

5

0.1016

4

0.6800

1

Hematite 17.527 17.691 16.719 20.863 10.124 13.206 16.401 18.222 7.5082 11.368

1

7.9603

1

8.1338

8

Apatite 23.027 23.284 21.285 27.148 15.386 18.522 21.57 23.944 2.7951 4.4297

9

2.9443

7

9.0051

7

Zircon 0.3685 0.3607 0.3463 0.4214 0.2475 0.3146 0.3408 0.378 0.0824 0.0912

5

0.0757 0.1114

4

Perovskite 0 0 0 0 0 0 0 0 0 0 0 0

Chromite 0.0023 0.0043 0.0025 0.0053 0.0079 0.0034 0.0042 0 0.0549 0.0543 0.0380

2

0.0304

9

Titanite 0 0 0 0 0 0 0 0 0 0 0 0

Pyrite 0 0 0 0 0 0 0 0 0 0 0 0

Halite 0 0 0 0 0 0 0 0 0 0 0 0

Fluorite 0 0 0 0 0 0 0 0 0 0 0 0

Anhydrite 0 0 0 0 0 0 0 0 0 0 0 0

Na2SO4 0 0 0 0 0 0 0 0 0 0 0 0

Calcite 30.913 22.557 30.602 18.189 64.488 55.153 29.197 22.57 43.138 18.533

8

39.484

5

40.498

9

Na2CO3 6.8315 8.7264 5.0699 8.0823 2.7393 4.6646 6.488 6.7865 16.33 23.348

9

19.875

3

13.926

7

159



Total 99.988 99.954 99.99 99.958 100.03 100.03 99.975 99.972 99.103 98.700

5

98.974

9

99.298

8

Fe3+/(Total

Fe) in rock

100 100 100 100 100 100 100 100 100 100 100 100

Mg/(Mg+Total

Fe) in rock

42.209 47.879 48.14 41.664 25.364 23.346 48.619 47.67 51.701 52.937

8

53.089

7

48.908

4

Mg/(Mg+Fe2+

) in rock

100 100 100 100 100 100 100 100 100 100 100 100

Mg/(Mg+Fe2+

) in silicates

100 100 100 100 100 100 100 100 100 100 100 100

Ca/(Ca+Na) in

rock

99.885 99.8 99.879 99.729 100 100 99.877 99.734 90.192 76.194

2

88.868

7

92.753

9

Plagioclase An

content

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

Differentiation

Index

1.1907 2.6196 2.6539 1.5184 0.3131 0.3949 2.6899 2.8899 15.063 21.874

5

15.229

9

14.605

5

Calculated

density, g/cc

3.209 3.2353 3.2047 3.323 2.9701 3.0384 3.1958 3.2583 2.8436 2.8964

8

2.8260

1

2.8834

Calculated

liquid density,

g/cc

2.8922 2.8921 2.8853 2.9146 2.8091 2.8379 2.8838 2.8941 2.7492 2.6891

3

2.7337

8

2.7600

7

Calculated

viscosity, dry,

Pas

-0.191 -0.158 -0.165 -0.177 -0.253 -0.249 -0.163 -0.157 -0.068 0.0480

1

-

0.0574

-0.084

Calculated

viscosity, wet,

Pas

-0.194 -0.168 -0.172 -0.184 -0.248 -0.244 -0.171 -0.166 -0.073 0.0371

3

-

0.0625

-

0.0951

Estimated

liquidus temp.,

°C

1920.1 1840.8 1856.7 1881.9 2074.7 2064 1853.7 1834.1 1695 1470.7

2

1679.2

3

1725.9

2

Estimated

H2O content,

wt. %

-3.454 -1.678 -1.875 -2.756 -11 -10.27 -1.808 -1.616 0.0985 0.1685

8

0.1089

9

0.0089

Normalized

analysis

SiO2 11.45 15.77 14.91 13.53 3.02 3.6 15.07 16.14 23.52 35.5 24.34 21.89

TiO2 1.4 1.42 1.31 1.65 0.77 0.96 1.29 1.46 0.19 0.31 0.25 0.31

Al2O3 2.39 2.65 2.18 3.09 1.1 1.34 2.21 2.57 1.6 2.46 1.67 1.77

Fe2O3 18.2 18.59 17.18 21.83 10.26 13.53 17 18.92 9.05 12.98 8.91 9.23

160



FeO 0 0 0 0 0 0 0 0 0 0 0 0

MnO 0.41 0.46 0.43 0.52 0.6 0.53 0.43 0.48 0.54 0.3 0.23 0.48

MgO 6.71 8.62 8.05 7.87 1.76 2.08 8.12 8.7 4.89 7.37 5.09 4.46

CaO 31.38 27.04 29.87 26.66 45.27 42.19 29.5 27.12 27.79 14.48 26.15 29.65

Na2O 0.02 0.03 0.02 0.04 0 0 0.02 0.04 1.67 2.5 1.81 1.28

K2O 0.01 0.02 0.01 0.02 0 0.01 0.01 0.02 0.07 0.12 0.09 0.2

P2O5 10.32 10.56 9.44 12.26 6.73 8.19 9.65 10.73 1.31 2.16 1.41 4.17

CO2 17.06 14.23 15.99 11.88 29.89 26.83 16.1 13.23 27.96 20.16 28.42 25.31

SO3 0 0 0 0 0 0 0 0 0 0 0 0

S 0 0 0 0 0 0 0 0 0 0 0 0

F 0 0 0 0 0 0 0 0 0 0 0 0

Cl 0 0 0 0 0 0 0 0 0 0 0 0

Sr 0.3597 0.27 0.3315 0.2725 0.3888 0.4695 0.3035 0.2575 0.3334 0.1775

1

0.4671

2

0.3444

9

Ba 0.0152 0.0311 0.0236 0.03 0.0668 0.0749 0.0266 0.0358 0.031 0.0136

2

0.0107

2

0.0710

1

Ni 0.0042 0.005 0.0042 0.0053 0.0017 0.002 0.0042 0.0048 0.0485 0.0584

1

0.042 0.0306

7

Cr 0.0016 0.0031 0.0018 0.0038 0.0054 0.0023 0.0029 0 0.0405 0.0416

5

0.0286

5

0.0222

2

Zr 0.2572 0.2548 0.2392 0.2964 0.1686 0.2167 0.2375 0.2638 0.0601 0.0693 0.0564

6

0.0803

7

Total 99.988 99.954 99.99 99.958 100.03 100.03 99.975 99.972 99.103 98.700

5

98.974

9

99.298

8

Ferric iron

selection

Normalized

analysis

total=100%?

n n n n n n n n n n n n

User Fe3+/FeT 1 1 1 1 1 1 1 1 1 1 1 1

Norm

calculation

check Value: OK

Norm seems OK

161

Figure 4.12. REE distribution patterns of carbonatite from Al-Uyaynah (this study) compared with

those previously published (Woolly. 1991)., and the global average of carbonatite. The values are

normalized to C1-chondrite [40]. The diagram (a) illustrates the REE distribution of ferro-carbonatite

that were collected from fields vs the Al-Uyaynah carbonatites of Woolley 1991 [39], and world

average of ferro-carbonatite. (b) shows the REE distribution of calcio-carbonatite and average global

and Woolley average of calcio-carbonatites.

1

10

100

1000

10000

La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu

C1

-ch

ond

rite

no

rmal

ized

Calcio-carbonatite

5 W. Al-Uyaynah calcio-carbonatite

World average calcio-ocarbonatite

1

10

100

1000

10000

100000

La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm Yb Lu

C1

-ch

ond

rite

no

rmal

ized

Ferro-carbonatite

1 2

4 27

28 W. Al-Uyaynah ferro-carbonatite

World average ferro-carbonatite

162

6. Discussion

The results of the combined geochemical and spectroscopy study on carbonatite shows that

carbonatite spectral behavior is governed by: 1) The bulk composition as reflected in

mineral/element concentrations; and 2) The presence or absence of a weathered surface. It is noted

that potential carbonatite heterogeneity may impact attempts to correlate spectral behaviors and

geochemically determined concentrations.

6.1. The relationship between bulk composition and carbonatite spectra

Overall results show that elemental concentrations of major and minor elements are

strongly correlated with the depth of the absorption features of carbonatite spectra. This correlation

holds for absorption features that are directly related to elemental vibrations (i.e. Nd3+, Fe), and

those where the geochemical composition is a proxy for a bond related absorption - CaO as a proxy

for CaCO3 and the C-O bond. On the basis of these spectral responses for carbonatites, the

characteristic mineralogy of the carbonatite may be simplified to calcite, iron-bearing phases

(oxides, chlorite, micas, amphibole, titanite, stilpnomelane), and REE-bearing phases (i.e. apatite,

Ce-allanite and Ce-monazite). The subsequent discussion focuses on elemental concentrations

rather than these minerals individually.

This geochemical analysis (XRF and ICP-MS) of the Al-Uyaynah carbonatite clearly

distinguish two groups (calcio-carbonatite and ferro-carbonatite), each with its own characteristic

spectra. Both carbonatite types contain differing proportions of calcite, iron-bearing phases, and

163

REE-bearing phases. Ferro-carbonatites contain a lesser amount of calcite and are more enriched

in iron-bearing and REE-bearing phases. In contrast, calcite is the most abundant mineral in calcio-

carbonatite, with a correspondingly lower concentration of iron-bearing phases and REE-bearing

phases. Results showed that the carbonatites' laboratory spectra were controlled by three key

minerals/elements:

1) Calcite concentration within carbonatites has a strong influence on the absorption depth

of the carbonatite spectra in SWIR area between 2.312–2.345 μm of the wavelength. This calcite

absorption was detectable in samples with calcite concentrations as low as 18% (calcite-poor ferro-

carbonatite). The calcio-carbonatite contains about ~55-64 % CaCO3 (per normalization) and

exhibit a greater absorption depth than the ferro-carbonatites. The correlation between the

concentration and wavelength absorption depth of our results show a positive linear correlation

with an R2 value about 0.8 (Figure 4.14).

2) Elemental iron (Fe2+, Fe3+), present in the form of an array of iron-bearing minerals

within the carbonatites, generated absorptions in the spectrum between 0.885-1.037μm. Our work

shows that the ASD instrument detected iron absorption over the full range of Fe2O3 concentrations

present within our suite (~9-21 wt. %). The concentration of iron-bearing minerals is greater within

the ferro-carbonatites compared to calcio-carbonatites, generating a more pronounced absorption

within the ferro-carbonatite samples. The iron absorption depth in carbonatites is well-correlated

with concentration (R2 > 0.8) (Figure 4.18). Complexities in the relationship between the iron-

absorption spectrum and the elemental concentration of Fe as reported by XRF analysis are

possible due to the variable oxidation state of iron. Carbonatites contain minerals where iron is

164

Figure 4.13. Calcite absorption center ranges measured from the carbonatite spectra.

Figure 4.14. Statistical comparison between calcite concentrations of carbonatites and the equivalent

laboratory spectra absorption depth records. The change in absorption depth due to the increase the

calcite concentrations of carbonatite samples spectra measured with the ASD in the laboratory. The R2

~ 0.8 (slope close to 1) was found among measured depth and chemically measured calcite

concentrations in weight percentage.

27_1

27

5

5_1

1

1_1

2

2_1

2828_1

44_2

y = 0.0059x - 0.0424

R² = 0.75520.000

0.050

0.100

0.150

0.200

0.250

0.300

0.350

0.400

0.450

0% 10% 20% 30% 40% 50% 60% 70%

Dep

th (

μm

)

Normative Calcite

concentration%

165

Figure 4.15. Displays the absorption center ranges of neodymium measured from the carbonatite

spectra.

Figure 4.16. Relationship between neodymium concentrations (ppm) and the spectra absorption depth

of carbonatites spectra. The neodymium absorption depths were calculated based on continuum removal

method of carbonatite spectra, and Nd concentrations were measured using, LA-ICP-MS. The

coefficient of determination of the Nd concentration vs its absorption depth displays a confident

correlation (R2 > 0.9).

27_1275 5_1

1 1_1

2

2_1

28

28_1

4

4_2

y = 2E-06x + 0.0009

R² = 0.9393

0.0000

0.0020

0.0040

0.0060

0.0080

0.0100

0.0120

0.0140

0.0160

0.0180

0.0200

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0

Dep

th (

μm

)

Nd concentration in thousands of PPM

166

Figure 4.17. Shows the absorption center ranges of iron oxides measured from the carbonatite spectra.

Figure 4.18. The iron oxides absorption feature depths of the carbonatite spectra were evaluated. The

regression relationships were used to understand the connection between iron oxides absorption depths

and its concentrations in carbonatites. The change in absorption depth with the increase of the iron

concentrations in carbonatites has good correlation (R2 > 0.8).

27_1

27

5

5_1

1

1_1

2

2_1

28

28_1

44_2

y = 0.0005x - 0.0011

R² = 0.8379

0.001

0.002

0.003

0.004

0.005

0.006

0.007

0.008

0.009

0.01

0.011

7% 9% 11% 13% 15% 17% 19% 21% 23%

Dep

th (

μm

)

Fe2O3 (wt.%)

167

Figure 4.19. Shows the absorption center ranges of Nd measured from the carbonatite spectra.

Figure 4.20. Nd concentration (in ppm) plotted against the spectra absorption depth of carbonatites from

this study and others [29]. The neodymium absorption depths were calculated based on continuum

removal method of carbonatite spectra. The coefficient of determination of Nd concentration vs its

absorption depth shows a positive correlation (R2 > 0.6).

CR-01

CR-03

CR-04

CR-14 CR-15

CR-17

CR-19

CR-20

CR-21

CR-24

CR-29

CR-31

CR-33

CR-37

CR-42

CR-43

CR-46

27_1

27

55_1

1

1_1

2

2_1

28

28_1

4

4_2

0

1000

2000

3000

4000

5000

6000

7000

8000

0.69 0.7 0.71 0.72 0.73 0.74 0.75 0.76 0.77 0.78

Nd

(p

pm

)

Nd absorption depth

CR-01

CR-03

CR-04

CR-14 CR-15

CR-17

CR-19 CR-20

CR-21

CR-24 CR-29

CR-31

CR-33

CR-37

CR-42 CR-43

CR-46 27_1

275 5_1

11_1

2

2_1

2828_1

44_2

R² = 0.61730

0.002

0.004

0.006

0.008

0.01

0.012

0.014

0.016

0.018

0.02

100 1000 10000

Nd

ab

sorp

tio

n d

epth

s

Nd (ppm)

168

Figure 4.21. Shows Nd and iron concentration (in ppm) plotted against spectra absorption depth. In the

samples with high REE-Nd concentration and low iron, the absorption is greater in depth.

743.2592.0

420.8

523.6

5467.1 3601.4

4441.3

6821.9

643.3 654.5 592.0 590.5

218300

171800

102600

135300

89100

92300

90500

129800

182000 185900170000 161800

0.0010

0.0100

Dep

th

(ppm)

Nd Fe

169

present in both Fe2+ (amphibole, mica, chlorite) and Fe3+ (iron-oxides). Given that XRF analysis

calculates all iron as Fe3+, the distribution of iron between the oxidation states from this

geochemical analysis is unknown. On the basis that Fe2+ (1 μm) and Fe3+ (0.9μm) have different

absorption peaks, interference between the peaks may result in imprecision in the combined iron

absorption depth.

3) The spectrum of REEs, which is dominated by Nd3+, is evident as an absorption feature

in the VNIR spectrum between 0.7 to 0.774 µm. Unsurprisingly the Nd3+ absorption depths for

carbonatites are more pronounced in samples with higher REE concentrations. An important

outcome from these results is, however, that Nd concentrations within a carbonatite must exceed

1000 ppm before it is detectable using ASD spectroscopic methods. The coefficient of

determination of the Nd concentration vs. its absorption depth exhibits a positive correlation (R2>

0.9 for Nd), (Figure 4.16).

6.2. The spectra of a carbonatite weathered surface

Remote sensing sensors (hyperspectral and multispectral) scan only the surface layer of

rock outcrops. Such weathered surface layers may exhibit significantly different spectral

signatures to the rock beneath. Carbonatite, in particular, is prone to surface degradation and

alteration of its surface. Despite the broad acknowledgement of the significant impact of this issue,

sample spectra collected in the lab frequently rely upon fresh cut surfaces. Utilizing these

unrepresentative spectra to interpret remote sensing images during carbonatite exploration or

mapping, could introduce an unconstrained source of ambiguity.

170

This study examined the spectroscopic and geochemical differences between the fresh

interior and weathered exterior of both ferro- and calico-carbonatites. There was a consistent

increase in the abundance of iron and REEs on the weathered exterior in comparison to the fresh

interior in both ferro- and calcio-carbonatites (Table 4.2). Accordingly, the iron and Nd3+

absorption depths were typically greater in the weathered exterior spectra in comparison to the

fresh interiors. In contrast, the abundance of calcite in the fresh interior was higher than the

weathered exteriors in all carbonatites tested. Therefore, the calcite absorption depths were greater

in the fresh interior spectra in comparison to the weathered exteriors. More importantly, the

particular features of the weathered exterior of carbonatites (high in iron, and REEs) may actually

result in making carbonatite more detectable using field spectroscopy instruments like ASD, and

hyperspectral remote sensing. While there was a lower calcite content in the weathered exterior, it

was still present within the spectra. These results suggest that the weathered exterior of carbonatite

might serve as a better remote sensing target because of the higher concentration of REEs and iron,

combined with evidence of a C-O bond from the remaining calcite. Thus, while the weathered

surface may enhance carbonatite detection, ambiguity may remain as to the actual content of REE

because of elemental redistribution.

6.3. Nd spectra absorption features in carbonatite rocks

The application of spectroscopic techniques to characterize carbonatite rocks is

uncommon. To our knowledge, such work has been undertaken in only few other studies. That

study examined a wider array of global carbonatites[29], and did not consider highly constrained

samples such as those presented herein . The results of this prior study are broadly similar to our

171

results and thus the datasets are examined in tandem. There are, however, differences among the

datasets: 1) the previously published spectra were collected on a different aliquot of rock to that

used for geochemical analysis. Indeed, the geochemical data is from work published from multiple

studies and labs. Carbonatite is known to have potentially extreme heterogeneity in terms of REE

distribution due to the concentration of REEs in particular phases. Accumulations of coarse

allanite, monazite, or apatite, for example, in one part of the rock that may have been analyzed

geochemically may lead to difficulties when correlated with a spectral signal taken on a surface

from another part of the sample lacking such accumulations. 2) Prior studies examined only fresh

interior surfaces of the carbonatite, limiting the application of such spectra in real world conditions

where field outcrops are dominated by weathered surfaces. Furthermore, it was not possible to

apply the full range of parallel transformations to these spectral results, as the published data were

not in a digital format.

Our results are comparable with earlier examinations, showing that the Nd3+ absorption

center was found at ~0.744 μm. When comparing the combined Nd concentrations (data from this

study and the previously published data) with the spectral absorption depths, it resulted in the R2

values that are > 0.6 (Figure 4.20). Despite the poorness of the linear fit of the data, the results do

show a rough positive relationship between the Nd concentrations and the spectra absorption depth

(Figure 4.18, 4.20, and 4.21).

Difficulties in measuring the relationship between Nd concentration and Nd3+ absorption

depths are further complicated by the impact of iron in terms of spectral interference. Studies which

examined artificial sample mixtures (iron + Nd) showed that when iron oxide reaches more than

172

10%, it significantly hinders REE-related absorption depth [37]. The impact of this interference

has been noted and indeed some of the scatter evident in our combined dataset may arise from the

high concentration of iron in some samples. The impact of iron interference for samples from our

study is potentially significant give Fe2O3(T) concentrations in our samples are ~9-21 wt.%, which

may cause REEs absorption depth reduction (figure 4.21).

7. Conclusion

Our interpretation of the normalized carbonatite sample spectra, prepared using the

continuum removal method, is that carbonatite minerals and elemental concentrations are

generally characterized by multiple diagnostic absorption features within the VNIR and SWIR

wavelengths. However, these characteristic absorption bands are variable between different types

of carbonatites. The spectroscopic outcomes of this investigation show that three key

minerals’/elements’ (calcite, iron oxides, and REE-Nd) absorptions were very distinctive in their

reflectance spectra, demonstrating a direct connection between carbonatite spectra and the

geochemical results. The concentration of calcite in carbonatites has a substantial effect on the

absorption depth between 2.312–2.345 μm of carbonatite spectra. The deepest absorption was

recorded in calcio-carbonatite. The correlation between the calcite concentration and wavelength

absorption depth shows a positive linear correlation with an R2 value about 0.8. The absorption

features between 0.885-1.037μm of the wavelength are caused by iron (Fe2+, Fe3+), which occurs

within iron-bearing minerals of carbonatites. The iron absorption was detected in all our samples

and the absorption depths were greater in ferro-carbonatite samples than in calcio-carbonatite

samples. The absorption depth of iron in our samples is well-correlated with concentration (R2 >

173

0.8). The REE absorption features that are governed by Nd3+ are recorded in VNIR area between

0.7 to 0.774 µm. In the samples with high REE-Nd concentration, the absorption depth is greater.

The Nd concentration vs. its absorption depth scatterplots diagram shows a positive correlation

with R2> 0.9.

Presented here is the first comparison of the impact of the development of a weathered

surface on carbonatite spectra. The weathered exterior consistently exhibits increased

concentrations of iron and REE, with a corresponding drop in carbonate concentrations. These

geochemical variations were evident within the rock spectra, with more pronounced absorptions

on weathered surfaces. These results suggest that weathered carbonatite surfaces may be easier to

detect using local spectroscopic methods or in remote hyperspectral sensing images of

carbonatites. It is also note that care must be taken when predicting potential REE resources from

the magnitude of these spectra given that the weathered surface is thus not representative of the

whole rock concentration of REEs.

The Nd spectral absorption depth from this study was compared with available published

dataset. The results were compatible with earlier analyses. Most Nd absorption centers were

measured around 0.744 μm. The scatterplot diagram shows that the samples with Nd

concentrations more than 1000 ppm are more distinguishable in laboratory spectra than samples

with less Nd abundance. The high concentration of iron in some samples reduced the Nd absorption

depth, in particular for samples with Fe2O3 more than 13 wt.%. All Nd concentrations were plotted

against the spectral absorption depths and the R2 values are > 0.6. However, despite the potential

for carbonatite heterogeneity, the results show positive correlation.

174

While this study has helped to constrain carbonatite spectroscopy, more investigation is

needed to enhanced interpretations of REE spectra. In particular, focused research on the

relationship between iron concentration and Nd spectra are needed to fully interpret spectral

results.

175

REFERENCES

176

REFERENCES

[1] P. L. Verplanck and B. S. Van Gosen, "Carbonatite and alkaline intrusion-related rare earth

element deposits–A deposit model," US Geological Survey 2331-1258, 2011.




"Extrusive Carbonatites from the Uyaynah Area, United Arab Emirates," J. Petrology, vol.

32, pp. 1143-1167, December 1, 1991 1991.

[4] R. Clark, "Spectroscopy and principles of spectroscopy, manual of remote sensing," John

Wiley and Sons, Inc, 1999.

[5] C. Elachi and J. J. Van Zyl, Introduction to the physics and techniques of remote sensing

vol. 28: John Wiley & Sons, 2006.

[6] R. N. Clark, "Spectroscopy of rocks and minerals, and principles of spectroscopy," Manual

of remote sensing, vol. 3, pp. 3-58, 1999.

[7] S. J. Gaffey, "SPECTRAL REFLECTANCE OF CARBONATE MINERALS IN THE

VISIBLE AND NEAR INFRARED (0.35-2.55 um).'," Journal of Geophysical Research,

vol. 92, pp. 1429-1440, 1987.

[8] R. P. Gupta, Remote sensing geology: Springer Science & Business Media, 2013.


minerals and rocks; IV, Sulphides and sulphates," Modern Geology, vol. 3, pp. 1-14, 1971.

[10] J. B. Adams, "Visible and near‐infrared diffuse reflectance spectra of pyroxenes as applied

to remote sensing of solid objects in the solar system," Journal of Geophysical Research,

vol. 79, pp. 4829-4836, 1974.

[11] R. Burns, "Origin of electronic spectra of minerals in the visible to near-infrared region,"

Remote geochemical analysis: Elemental and mineralogical composition, vol. 4, pp. 3-29,

1993.

[12] F. Van Der Meer, "Analysis of spectral absorption features in hyperspectral imagery,"

International journal of applied earth observation and geoinformation, vol. 5, pp. 55-68,

2004.

177

[13] T. A. Warner, G. M. Foody, and M. D. Nellis, The SAGE handbook of remote sensing:

Sage Publications, 2009.

[14] G. R. Hunt, "Electromagnetic radiation: the communication link in remote sensing,"

Remote sensing in geology, vol. 2, pp. 5-45, 1980.

[15] A. F. Goetz, "Spectral remote sensing in geology," Theory and Applications of Optical

Remote Sensing,[1989], Inc, pp. 491-52, 1989.

[16] R. N. Clark, Spectroscopy of rocks and minerals and principles of spectroscopy. United

States (USA): John Wiley & Sons, New York, NY, United States (USA), 1999.

[17] A. Kerr, H. Rafuse, G. Sparkes, J. Hinchey, and H. Sandeman, "Visible/infrared

spectroscopy (VIRS) as a research tool in economic geology: background and pilot studies

from NewFoundland and Labrador," Geological Survey, Report, vol. 11, pp. 145-166,

2011.

[18] G. R. Hunt, "Spectral signatures of particulate minerals in the visible and near infrared,"

Geophysics, vol. 42, pp. 501-513, 1977.

[19] L. C. Rowan and J. C. Mars, "Lithologic mapping in the Mountain Pass, California area

using advanced spaceborne thermal emission and reflection radiometer (ASTER) data,"

Remote sensing of Environment, vol. 84, pp. 350-366, 2003.

[20] R. N. Clark, T. V. King, M. Klejwa, G. A. Swayze, and N. Vergo, "High spectral resolution

reflectance spectroscopy of minerals," Journal of Geophysical Research: Solid Earth

(1978–2012), vol. 95, pp. 12653-12680, 1990.

[21] R. N. Clark, G. A. Swayze, R. Wise, K. E. Livo, T. M. Hoefen, R. F. Kokaly, et al., USGS

digital spectral library splib06a: US Geological Survey Reston, VA, 2007.



[23] F. Van der Meer, "Spectral reflectance of carbonate mineral mixtures and bidirectional

reflectance theory: Quantitative analysis techniques for application in remote sensing,"

Remote Sensing Reviews, vol. 13, pp. 67-94, 1995/08/01 1995.

[24] R. N. Clark, G. A. Swayze, K. E. Livo, R. F. Kokaly, S. J. Sutley, J. B. Dalton, et al.,

"Imaging spectroscopy: Earth and planetary remote sensing with the USGS Tetracorder

and expert systems," Journal of Geophysical Research: Planets (1991–2012), vol. 108,

2003.

178

[25] F. D. Van der Meer, H. M. Van der Werff, F. J. van Ruitenbeek, C. A. Hecker, W. H.

Bakker, M. F. Noomen, et al., "Multi-and hyperspectral geologic remote sensing: A

review," International Journal of Applied Earth Observation and Geoinformation, vol. 14,

pp. 112-128, 2012.

[26] L. C. Rowan, M. J. Kingston, and J. K. Crowley, "Spectral reflectance of carbonatites and

related alkalic igneous rocks; selected samples from four North American localities,"

Economic Geology, vol. 81, pp. 857-871, 1986.

[27] T. L. Bowers and L. C. Rowan, "Remote mineralogic and lithologic mapping of the ice

river alkaline complex, British Columbia, Canada, using AVIRIS data," Photogrammetric

Engineering and Remote Sensing, vol. 62, pp. 1379-1386, 1996.


Greenland, using HyMap imaging spectrometer data," Remote Sensing of Environment,

vol. 113, pp. 1208-1219, 2009.

[29] D. A. Neave, M. Black, T. R. Riley, S. A. Gibson, G. Ferrier, F. Wall, et al., "On the

feasibility of imaging carbonatite-hosted rare earth element (REE) deposits using remote

sensing," 2016.

[30] R. N. Clark and T. L. Roush, "Reflectance spectroscopy: Quantitative analysis techniques

for remote sensing applications," Journal of Geophysical Research: Solid Earth, vol. 89,

pp. 6329-6340, 1984.

[31] M. F. Noomen, A. K. Skidmore, F. D. Van der Meer, and H. H. Prins, "Continuum removed

band depth analysis for detecting the effects of natural gas, methane and ethane on maize

reflectance," Remote Sensing of Environment, vol. 105, pp. 262-270, 2006.

[32] R. F. Kokaly, "PRISM: Processing routines in IDL for spectroscopic measurements

(installation manual and user's guide, version 1.0)," US Geological Survey 2331-1258,

2011.



[34] T. O. Rooney, W. K. Hart, C. M. Hall, D. Ayalew, M. S. Ghiorso, P. Hidalgo, et al.,

"Peralkaline magma evolution and the tephra record in the Ethiopian Rift," Contributions

to Mineralogy and Petrology, vol. 164, pp. 407-426, 2012.

[35] T. O. Rooney, K. D. Morell, P. Hidalgo, and P. Fraceschi, "Magmatic consequences of the

transition from orthogonal to oblique subduction in Panama," Geochemistry, Geophysics,

Geosystems, vol. 16, pp. 4178-4208, 2015.


179

[36] R. N. Clark, G. A. Swayze, R. Wise, K. E. Livo, T. M. Hoefen, R. F. Kokaly, et al., "USGS

digital spectral library splib06a," ed: US Geological Survey Reston, VA, 2007.

[37] N. K. Boesche, C. Rogass, C. Lubitz, M. Brell, S. Herrmann, C. Mielke, et al.,

"Hyperspectral REE (Rare Earth Element) Mapping of Outcrops—Applications for

Neodymium Detection," Remote Sensing, vol. 7, pp. 5160-5186, 2015.

[38] R. L. Cullers and J. L. Graf, "Rare earth elements in igneous rocks of the continental crust:

predominantly basic and ultrabasic rocks," in Rare earth element geochemistry. vol. 2, ed:

Elsevier Amsterdam, 1984, pp. 237-274.

[39] K. Bell, "Carbonatites : genesis and evolution," London.

[40] S.-S. Sun and W. McDonough, "Chemical and isotopic systematics of oceanic basalts:

implications for mantle composition and processes," Geological Society, London, Special


180

5. CHAPTER 5

CONCLUSION AND RECOMMENDATIONS

1. Dissertation Synthesis

The main objectives of this dissertation are to explore remote sensing data, mapping

algorithm techniques, and spectroscopic applications to identify and map carbonatite distribution

at Al-Uyaynah area-UAE. The research surveys carbonatites by employing the Visible/Near Infra-

Red (VNIR) and Short Wave Infra-Red (SWIR) with three datasets: (1) the latest multispectral

sensor (Landsat 8 OLI), (2) ASTER imaging and (3) Analytical Spectral Device (FieldSpec-3 Hi-

Res ASD) Spectroradiometer. Whole rock geochemical analysis of major element oxide and REE

concentrations in carbonatites were determined using Standardized X-ray fluorescence (XRF) and

Laser Ablation Inductively Coupled Plasma Mass Spectrometry (LA-ICP-MS). Geographic

distribution of carbonatites was outlined (chapter 2) by an examination of Landsat 8 OLI data

validation, as it is considered to be one of the most widely accessed open-data sources used for

geological mapping. This presents a novel approach of exploring the Al-Uyaynah area, utilizing

new Landsat 8 imagery through mapping algorithms. Following this investigation, carbonatite

distribution in Al-Uyaynah area was examined utilizing ASTER data processed with four various

mapping processing algorithms: the Binary Encoding Classification (BEC), Spectral Information

Divergence (SID), Spectral Angle Mapper (SAM), and Support Vector Machine (SVM) (Chapter

3). These mapping algorithms were employed to analyze and digitally classify carbonatites and

181

spatially adjunct geological units with the tested outcomes ranked based on their accuracies. The

third investigation explores carbonatite chemistry using spectral analysis devices (ASD) and

geochemical approaches to develop a comprehensive understanding of chemical environments of

the Al-Uyaynah carbonatites. These geochemical data allow for testing of the relationship between

the major oxides and REEs spectral absorption features of carbonatite spectra and their

concentrations in carbonatites linearly (chapter 4). The overall findings of all dissertation chapters

are summarized in this section, employing this information to assess the effectiveness of Landsat

8, ASTER, Analytical Spectral Device (ASD) data for carbonatites mapping in Al-Uyaynah area.

Moreover, the broader impacts concerning the utility of these datasets and the tested mapping

algorithms in future exploration will also be addressed.

2. Dissertation conclusion

Remote sensing imagery has been a standard tool for exploration and mineral mapping

over the past four decades, initially utilized as a rapid and cost-effective approach to analyzing the

chemical variability of surface rock and mapping of lithological units [8, 9]. In this research, I

have analyzed two multispectral satellite systems in terms of their utility for carbonatite indication

and mapping; the newest generation of Landsat satellites (Landsat 8 OLI) and ASTER visible/near-

infrared (VNIR); and Short Wave Infrared (SWIR) data. Both Landsat and ASTER imagery

remains the most popular open-source remote sensing data and are widely used by the geologic

and remote sensing communities. The performances of both systems were compared for use in

carbonatite mapping, and an assessment of these systems validates their efficacy in the mapping

of carbonatite units. ASTER outperformed the other systems, suggesting that spectral resolution

182

is more crucial when it comes to carbonatite mapping, as it provides more spectral bands in SWIR

to detect carbonatite key minerals.

Of the many image categorization techniques that rely on the spectral method, we

examined four mapping methods (BEC, SID, SAM, and SVM) to determine the best method to

detect and map carbonatite and other associated rock units within the study area. The classification

method maps were compared with the field ground truth data, resulting in variability among their

results. The best discrimination between the image's end-members was achieved using the SVM

algorithm, producing 86% (excellent) overall agreement with the field ground's truth and the

current published geological map. The SAM algorithm provided a good map with 61% accuracy

whereas the other two classifiers (BEC, SID) performed poorly. Despite the excellent classification

that was obtained using the SVM learning algorithm, some degree of classification error was

observed: i) Carbonatite complexes generally appear in fields as small units within larger-scale

features, complicating the mapping process. ii) The similar richness in iron oxide minerals in the

weathered outcrop of ferro- and calcio-carbonatite layers resulted in difficulty in separating the

units. iii) The mapping of ferro-carbonatites using the SVM algorithm was more challenging in

the southern region because they are narrow and structurally deformed units that occur near the

alkali volcanic rocks. The blended appearance of the rock units makes it difficult to separate the

units using the SVM algorithm. Regardless of this difficulty, our overall experiment results suggest

the SVM algorithm is useful for mapping carbonatites using remote sensing data.

One of the objectives of this research was to analyze the spectral features carbonatites

utilizing field spectraradiometer of chemically analyzed rocks. In this regard, we investigated

183

whether the reflectance spectroscopy, in the area between to the 0.45 to 2.5 μm, can be utilized to

identify and detect carbonatite minerals/elements. Spectra derived using the continuum removal

method show multiple diagnostic absorption features within VNIR and SWIR bands. Specifically,

the investigation shows three key minerals/elements leading to very distinctive absorptions in the

reflectance spectra (calcite, iron, and REE-Nd). Comparing the spectra of the carbonatites to the

geochemical analyses demonstrates positive correlation between the observed absorption depth

and the carbonatite key minerals/elements abundances leading to a promising novel approach to

more quickly and at lower cost distinguish between types of carbonatite. The presence of VNIR

spectral features resulting from REE-Nd3+ make it possible to identify potential REE ore deposits

for the purpose of prospecting. This research outcome adds to data on carbonatite spectroscopy,

which previously was limited or lacking. Moreover, some of the limitations of satellite system data

can be overcome by laboratory or field reflectance spectroscopy.

3. Recommendations and outlooks for carbonatite prospectors

Digital mapping based on remote sensing dataset and algorithm techniques, like those

previously reported within this dissertation, will not eliminate the necessity for the traditional field

based geological mapping. Rather, digital mapping offers the most cost-effective and time-saving

technique for generating maps that can be used as a guide for field investigation. This is

particularly valid when it comes to large scale area mapping. Given the spatial association of

carbonatites with silicate rocks that share some spectral features with carbonatites, such field

investigations remain a critical component in prospecting.

184

Mapping algorithms are many and produce different results depending on how the

algorithms manage image data. Depending on the algorithms efficiency and the quality of remote

sensing dataset, a resulting carbonatite map can be poor, or highly accurate. High accuracy maps

will allow researchers to focus their field investigation. In this research, we tested four algorithms

to map carbonatite units. The highest mapping accuracy was accomplished utilizing the SVM

algorithm. This algorithm illustrates the advantage of artificial intelligence or support vector

machine learning algorithm (SVM) over strictly static algorithms used in this research (BEC, SID,

SAM). Additional examination is required using high-resolution sensors to determine whether the

overall carbonatite mapping accuracy could be increased by using the SVM learning algorithm to

produce object-based carbonatite mapping.

Geospatial researchers are currently enjoying one of the most fertile periods for Earth

observing data. The remote sensing sensors are scanning the earth's surface at ever finer spatial

resolutions. The continuing decline of the expense of these data will help researchers pose new

questions and challenges that will result in the advancement of carbonatite identification and

geological mapping in general.

remote sensing applications for carbonatite …€¦ · carbonatites, alkali volcanics,...

Documents